Potential application of Healitide-GP1, a novel antibacterial peptide, in wound healing: in vitro studies

Software and tools

All computational analyses were performed using the following software:

BioPython (v1.79): Sequence manipulation and feature extraction.

ProPy (v1.1.2): Physicochemical property calculations.

PeptideDescriptor (v0.9.1): Feature generation.

Scikit-learn (v1.0.2): Machine learning model training.

PyCaret (v2.3.10): Model comparison and selection.

LightGBM (v3.3.2): Final classification model.

BLAST+ (v2.12.0): Sequence similarity searches.

I-TASSER (v5.1): Protein structure prediction.

HeliQuest (v2.0): Helical wheel projections.

MEGA X (v10.2.6): Phylogenetic analysis.

EMBOSS Needle (v6.6.0): Sequence alignment.

AMPDiscover (v1.1.0): Secondary validation of antimicrobial activity.

Dataset curation

WHP dataset

The WHP dataset was curated by UniProt by restricting specific search terms such as ‘wound healing’, ‘wound repair’ and ‘tissue repair’ to reviewed entries (e.g. UniProt query: ‘reviewed: true AND (wound healing OR wound repair OR tissue repair)’). After applying philtres (sequence length 4–100 amino acids, removal of sequences with ambiguous characters), 1000 unique WHPs were selected. Redundancy was controlled using CD-HIT with an identity threshold of 90. The non-WHP dataset was similarly curated from UniProt and Antimicrobial Peptide Database (APD), ensuring no overlap with WHP annotations, and processed with the same filters.

To develop WHPs, two datasets were created: the WHP dataset and the non-WHP dataset. The WHP dataset, containing known WHPs, was collected from publicly available databases such as UniProt, focusing on selecting “vetted’ entries to ensure high quality of manually curated data, and downloaded in FASTA format. The non-WHP dataset, which served as a negative example, was retrieved from databases such as UniProt and APD and downloaded in FASTA format. Finally, the WHP and non-WHP datasets were combined to train and evaluate the machine learning models.

Dataset preprocessing

To ensure the quality and relevance of the data for model training, several pre-processing steps were applied to the peptide sequences. First, the sequences were filtered to include only those between 4 and 100 amino acids in length, as peptides outside this range are less likely to have therapeutic activity. Sequences containing pseudo-amino acids or ambiguous characters were removed to maintain data integrity. In addition, duplicate sequences were removed to ensure the uniqueness of each entry in the dataset. After these pre-processing steps were completed, the resulting dataset was divided into a training (70%) and a test group (30%), which were used for the subsequent phases of model development and evaluation.

Feature extraction and selection

To characterize the potential wound-healing properties of the peptides, key features were extracted, encompassing various aspects of their sequence and physicochemical characteristics. These included the Amino Acid Composition (ACC), representing the proportion of each individual amino acid, and the Dipeptide Composition (DPC), capturing the frequency of consecutive amino acid pairs. Furthermore, crucial physicochemical properties such as amphiphilicity, hydrophobicity, net charge, and isoelectric point. Additional features like sequence length and molecular weight were also included. Feature extraction was performed using the Python libraries BioPython, ProPy, and PeptideDescriptor. Finally, to improve model efficiency and focus on the most informative features, correlation analysis was employed to identify and remove redundant features, retaining only features with significant predictive power.

Model development

To classify the peptides as either wound-healing or non-wound-healing based on the extracted features, several machine learning algorithms were trained and compared. These included the Gradient Boosting Classifier (GBC), which builds a strong predictive model by combining multiple weak classifiers like decision trees and is optimized through hyperparameter tuning. In addition, the XGBoost algorithm, which is a highly efficient gradient-boosting algorithm known for its ability to handle large datasets and support custom loss functions, was implemented. Finally, LightGBM, another gradient-boosting algorithm optimized for speed and memory efficiency through techniques like gradient-based one-side sampling (GOSS) and exclusive feature bundling (EFB), was also utilized.

The implementation of all models was performed using the Scikit-learn and PyCaret libraries, with PyCaret facilitating the comparison of performance across the different algorithms. The model performance was rigorously evaluated using several key metrics: Accuracy, representing the percentage of correct predictions; F1 score, which provides the harmonic mean of precision and recall; AUC-ROC, measuring the area under the receiver operating characteristic curve; and the Matthews Correlation Coefficient (MCC), a balanced measure of prediction quality that considers true and false positives and negatives. The model exhibiting the best performance across these metrics was selected for subsequent analysis and the generation of novel WHPs.

Genetic algorithm for peptide optimization

The genetic algorithm (GA) was used to generate potential wound-healing peptides (GP-WHPs). The algorithm begins with an initial population of peptides sourced from the WHP dataset. The fitness of each peptide was then assessed on the basis of the probability (as predicted by the best-performing machine learning model) of it being a WHP. Peptides exhibiting a probability score exceeding 80% were selected as the parents for the subsequent generation. Crossover operations were performed to combine the genetic material from these parent peptides, and random mutations were introduced to enhance the diversity of the offspring. Constraints were enforced throughout this process to ensure that the generated peptides remained within a length range of 4–100 amino acids. The GA iteratively proceeded for a maximum of 5 generations, or until the offspring demonstrated significant divergence from their parents while concurrently maintaining high fitness scores, indicating the potential generation of novel and effective WHPs.

Reverse search

To validate GP-WHPs and assess their novelty, a reverse search approach was employed. This involved using the BLASTp tool to search for GP-WHP sequences in the NCBI nonredundant protein database. The study identified homologous sequences in the database. By setting appropriate thresholds for sequence coverage and identity during the BLASTp search, it was possible to distinguish between potentially novel GP-WHPs and those with significant similarity to known proteins, thus providing an initial indication of their potential wound-healing properties and uniqueness.

Validation of the peptides

The generated peptides were further validated using a multi-faceted approach. First, a secondary machine learning validation was performed using online tools like AMPDiscover to re-evaluate the predicted wound-healing properties of the GP-WHPs, providing an independent assessment of their potential. Second, a detailed analysis of key physicochemical properties was conducted, focusing on hydrophobicity, net charge, and amphipathicity, which are known to influence peptide activity and stability. Finally, to assess the structural relationship between the generated peptides and known WHPs, pairwise sequence alignment was performed using tools like EMBOSS Needle, allowing for the identification of similarities and potential conserved regions that might contribute to wound-healing functionality.

Feature correlation analysis

To visualize the relationships between the extracted features and identify the key attributes contributing to the wound-healing activity, a correlation heatmap was generated. This heatmap provides a visual representation of the pairwise correlations between different features, allowing for the identification of trends and potential synergistic or antagonistic relationships among them.

Bioinformatics analysis

After peptide design, comparative analysis and phylogenetic study of a novel wound-healing peptide designed using deep learning was compared with experimentally validated wound-healing peptides to evaluate its potential and analyze its relationship with known peptides. Multiple sequence alignment (MSA) was performed to identify similarities and conserved regions among the peptides. Furthermore, a phylogenetic tree was constructed to determine the evolutionary relationships and classify the designed peptide relative to known peptides. Through this comparative analysis, insights into the structural and functional similarities of the novel peptide with well-established wound-healing peptides were obtained, enabling further assessment of its therapeutic potential. The amphipathic nature of the designed peptide was analyzed using helical wheel projection. The amino acid sequence was input into the HeliQuest software, which generated a graphical representation of the peptide’s structure assuming an alpha-helical conformation. This projection visualizes the arrangement of the amino acid residues around the helix, revealing the distribution of the hydrophobic and hydrophilic residues, thereby highlighting the amphipathic character of the peptide. To better understand its structure and function, we employed the I-TASSER server to predict its three-dimensional structure and identify potential functional sites.

Peptide synthesis and functional assays

Peptide synthesis

Healitide-GP1 was synthesized using solid phase peptide synthesis (SPPS) on an automated peptide synthesizer (Liberty Blue, CEM Corporation, USA). Fmoc chemistry (9-fluorenylmethoxycarbonyl) with Rink amide resin (0.49 mmol/g, 100–200 mesh) as a solid support was used in the synthesis. Each amino acid coupling step was performed with HBTU/HOBT activation in DMF solvent (20 min coupling time). After completion of the synthesis, the peptide was cleaved from the resin with a cleavage cocktail (TFA/water/TIS 95:2.5:2.5) for 2 h at room temperature. The crude peptide was precipitated with cold diethyl ether, collected by centrifugation (4000 rpm, 10 min) and lyophilized. The purity (96.8%) and molecular weight (2754.3 Da) were confirmed by analytical HPLC (Shimadzu Prominence, Japan) and mass spectrometry (Bruker Daltonics, USA).

Cell culture and evaluation of cytotoxicity

Human dermal fibroblasts (HDFs, Pasteur Institute Cell Bank, Iran) and human keratinocytes (HaCaT, Pasteur Institute Cell Bank, Iran) were cultured in DMEM (Gibco, USA) and RPMI-1640 medium (Gibco, USA), respectively, supplemented with 10% fetal bovine serum (FBS, Gibco, USA) and 1% penicillin/streptomycin (100 U/mL). The cells were maintained at 37 °C and 5% CO_a. For cytotoxicity assessment, cells were seeded in 96-well plates (5 × 10³ cells/well) and incubated with peptide concentrations (0–200 µg/mL) for 24 h (n = 6 replicates per concentration). Cell viability was determined using the MTT assay (Sigma-Aldrich, USA). In brief, 20 µL of MTT solution (5 mg/mL in PBS) was added to each well and incubated for 4 h. The formed formazan crystals were dissolved in DMSO (150 µL/well) and the absorbance was measured at 570 nm using a microplate reader (BioTek, USA). Viability was calculated as a percentage of untreated controls.

In vitro wound healing test (scratch assay)

HDFs and HaCaT cells were cultured to confluence in 24-well plates. A sterile pipette tip (200 µL) was used to create a standardized scratch wound. The wells were washed twice with PBS to remove detached cells. Fresh medium containing 50 µg/mL Healitide-GP1 was added (n = 6 replicates per cell type). Untreated wells served as controls. Wound closure was monitored after 0 and 24 h using an inverted microscope (Olympus CKX41, Japan) and a digital camera. The wound area was quantified using ImageJ software (v1.53k, NIH, USA) and the percentage wound closure was calculated as follows:

Antibacterial activity

Broth microdilution assays against Staphylococcus aureus (S.aureus) (ATCC 25923) and Escherichia coli (E.coli) (ATCC 25922) were performed to evaluate the antibacterial activity of the synthesized peptide. Serial dilutions of the peptide (from 0 to 200 µg/mL) were prepared in Mueller-Hinton broth. Bacterial growth was measured after 18 h of incubation at 37 °C by reading the optical density at 600 nm (OD₆₀₀₀). The minimum inhibitory concentration (MIC) and minimum bactericidal concentration (MBC) were determined using the broth microdilution method according to CLSI guidelines. Bacteria were cultured overnight in Mueller-Hinton broth (MHB) at 37 °C. Cultures were diluted to ~ 5 × 10⁵ CFU/ml in fresh MHB. Peptide solutions (0–200 µg/mL) were prepared in 96-well plates (100 µL/well). 100 µL of bacterial suspension was added to each well (final volume 200 µL). The plates were incubated for 18 h at 37 °C with shaking (200 rpm). The MIC was determined as the lowest concentration showing no visible growth. For MBC determination, 10 µL from MIC wells and higher concentrations were plated on MHA plates and incubated at 37 °C for 24 h. MBC was recorded as the lowest concentration with ≥ 99.9% kill (no colonies). All experiments were performed in triplicate technical design with biological duplicates (total n = 6).

Statistical analysis

Data are expressed as mean ± SD. Statistical significance was determined using Student’s t-test or one-way ANOVA with Tukey’s post-hoc test (GraphPad Prism v9.3.1, USA). P < 0.05 was considered significant.

Dataset preparation

The dataset was created to classify WHPs and non-WHPs. A search of the UniProt database yielded 12,453 unreviewed entries and 1,122 reviewed entries using relevant keywords for wound-healing activity. Only the reviewed entries were selected to ensure high confidence. After filtering, 1,000 WHPs were included in the dataset. To create a balanced dataset, 1,000 non-WHP sequences were randomly selected from unrelated peptide sequences available in UniProt and APD.

Following preprocessing, the final dataset comprised 2,000 peptide sequences (1,000 WHPs and 1,000 non-WHPs).

Feature extraction and selection

A total of 50 features were extracted from the peptide sequences:

ACC: e.g., the frequency of the C, K, and M residues.

DPC: e.g., GK, LL, CR.

Physicochemical properties: Hydrophobicity, net charge, isoelectric point, molecular weight, and amphipathicity.

After the correlation analysis, the top 10 most important features for classification were identified and ranked by feature importance in the LightGBM model. These features included amino acid composition (C, K, M), DPC (GK, LL, CR, FL, AG), hydrophobicity, and net charge.

Model selection and performance evaluation

Several machine learning models were trained on the dataset using PyCaret, which automatically split the data into 70% training and 30% test sets and performed 10-fold cross-validation. Table 1 lists the performance of the top 10 models.

The performance of the LightGBM model was evaluated using various metrics, including accuracy, precision, recall, F1 score and AUC-ROC. The model achieved an accuracy of 0.89, precision of 0.88, recall of 0.91 and F1 score of 0.89. The AUC-ROC value of the model was 0.90, as shown in Fig. 1. The hyperparameters of the LightGBM model were set as follows:

Learning rate: 0.1, Number of leaves: 31, Max depth: −1, Min child weight: 0.001, Subsample: 0.8, Colsample by tree: 0.8, Reg alpha: 0.01, Reg lambda: 0.01. These hyperparameters were tuned using a grid search approach to optimize the performance of the model.

The ROC curve and AUC for the LightGBM model are depicted in Fig. 1. The Receiver Operating Characteristic (ROC) curve provides a visual assessment of the LightGBM model’s performance in classifying wound-healing peptides (WHPs) versus non-wound-healing peptides (non-WHPs). On the x-axis, the False Positive Rate (FPR) indicates the proportion of non-WHPs that are incorrectly classified as WHPs. The y-axis represents the True Positive Rate (TPR), also referred to as sensitivity or recall, which corresponds to the proportion of WHPs correctly identified as WHPs. In the figure, the blue and orange curves represent the ROC performance for the WHP and non-WHP classes, respectively. The black dotted line indicates the baseline performance of a random classifier, where the TPR and FPR are equal. The Area under the Curve (AUC) serves as a comprehensive metric for evaluating the model’s classification ability. The LightGBM model achieved an AUC value of 0.90, reflecting a strong performance in differentiating between WHPs and non-WHPs. An AUC of 1.0 signifies a perfect classifier, while an AUC of 0.5 represents a random classifier. The ROC curve for the WHP class lies significantly above the diagonal baseline, illustrating the model’s robust discriminatory capability. Moreover, the curve’s proximity to the top-left corner indicates the model’s high sensitivity and specificity, meaning it accurately classifies a large percentage of WHPs (high TPR) while minimizing the incorrect classification of non-WHPs (low FPR). Overall, the ROC curves and corresponding AUC values demonstrate the effectiveness of the LightGBM model in distinguishing between wound-healing and non-wound-healing peptides. This strong performance supports the model’s potential as a valuable tool in the design and discovery of novel wound-healing peptides.

Generation of novel WHPs using a genetic algorithm

The genetic algorithm (GA) was applied to generate potential novel WHPs (GP-WHPs). Two rounds of GA were performed using the following constraints:

Round 1: No sequence length restriction.

Round 2: Sequence length limited to ≤ 100 amino acids.

The first round of the genetic algorithm generated 30 peptide sequences. After the BLASTp search and similarity analysis:

5 peptides had no significant similarity in the database and were considered to be potentially novel WHPs.

Ten peptides matched known wound-healing proteins.

The remaining sequences were annotated as antibacterial peptides or uncharacterized proteins (Table 2).

In round two, the peptide length was restricted to ≤ 100 amino acids (Table 3). This round generated 12 sequences, out of which: 4 sequences excluded the functional annotations. 3 sequences matched wound-healing proteins with high similarity. 2 sequences were identified as hypothetical proteins.

Validation of the WHP generated

The generated GP-WHPs were validated as follows:

1. 1.

   AMPDiscover for secondary QSAR prediction.

2. 2.

   Physicochemical properties (isoelectric point, hydrophobicity, net charge).

3. 3.

   BLASTp similarity search to confirm the novelty (Table 4).

Figure 2 illustrates the top 10 features contributing to the classification of wound-healing peptides (WHPs) versus non-WHPs, as determined by their importance scores using the LightGBM model. Feature importance, in this context, quantifies the relative contribution of each feature to the model’s predictive accuracy.

Dominance of Amino Acid Composition (C, K, M): The plot reveals that the amino acid composition, particularly the frequency of Cysteine (C), Lysine (K), and Methionine (M), plays a crucial role in distinguishing WHPs. Cysteine (C) is the most influential feature, suggesting its significant involvement in the structural and functional characteristics of WHPs. Lysine (K), known for its positive charge and role in various biological processes, ranks second, while Methionine (M), which is often involved in protein initiation and antioxidant activity, is the third most important feature.

Importance of DPC (GK, LL, CR): DPC also significantly contributes to the classification. The dipeptides Glycine-Lysine (GK), Leucine-Leucine (LL), and Cysteine-Arginine (CR) were identified as important discriminators. This result highlights the relevance of specific amino acid pairings for determining the wound-healing potential. For instance, the presence of GK may indicate specific structural motifs or interaction patterns characteristic of WHPs.

Physicochemical Properties Matter: Hydrophobicity, a key physicochemical property, is also among the top five features. Its presence in the top 10 underscores the importance of the peptide’s overall hydrophobic character in influencing its wound-healing activity.

Other Notable Features: The DPC involving Phenylalanine-Leucine (FL), the dipeptide Alanine-Glycine, and the net charge are also relevant to the model.

Net charge: The net charge is slightly less than the AG fee.

Implications:

The prominence of C, K, and specific dipeptides suggests that these amino acids and their combinations are strongly associated with the mechanisms underlying wound healing. Cysteine’s high importance may be attributed to its distinct chemical properties, which could contribute to the peptide’s stability, bioactivity, or interactions with biological molecules, independent of disulfide bond formation. The importance of positively charged residues like Lysine (K), Histidine (H), and Arginine (R) suggests that electrostatic interactions may play a role in WHP activity, perhaps through interactions with negatively charged cell membranes or other biomolecules involved in the wound-healing process.

The importance of hydrophobicity suggests that correctly identifying this parameter will facilitate better classification.

Overall, this feature importance analysis provides valuable insights into the molecular characteristics that differentiate WHPs from non-WHPs. These findings can guide the design and optimization of novel wound-healing peptides by focusing on the enrichment or manipulation of these key features.”

The correlation heatmap in Fig. 3 illustrates the relationships between the top 10 features identified by the LightGBM model for classifying wound-healing peptides (WHPs) and non-WHPs. The heatmap is color-coded, with red indicating positive correlations and blue indicating negative correlations.

Amino acid composition: Features related to amino acid composition, such as C, K, and M, are strongly correlated with each other. For example, C and K have a positive correlation of 0.86, whereas C and M have a negative correlation of −0.81. This suggests that the presence of certain amino acids is associated with the wound-healing ability.

DPC: Features related to the DPC, such as GK, LL, and CR, are strongly correlated with each other. For example, GK and LL have a positive correlation of 0.85, whereas GK and CR have a negative correlation of −0.78. This suggests that the presence of certain dipeptide combinations is associated with the wound-healing activity.

Physicochemical properties: Features related to physicochemical properties, such as hydrophobicity and net charge, show weaker correlations with each other. For example, the hydrophobicity and net charge have a positive correlation of 0.42. This suggests that the physicochemical properties of the peptides are less important than the amino acid and DPCs for the wound-healing activity.

Feature importance: The heatmap also shows the importance of each feature in wound healing. The features with the highest importance are C, K, and M, which are all related to the amino acid composition. This suggests that the presence of certain amino acids is critical for the wound-healing ability.

Overall, the correlation heatmap provides valuable insights into the relationships between the features identified by the LightGBM model and the wound-healing activity. The results suggest that the amino acid and DPCs of the peptides are the most important factors for wound-healing activity, and that certain physicochemical properties may also play a role.

The selection of Healitide-GP1 (GP-WHP1) as the lead candidate was based on a comprehensive analysis of its key properties and its performance in various bioassays. While GP-WHP8 had a higher QSAR value of 0.98, Healitide-GP1 showed a unique combination of hydrophobicity, net charge and bioactivity that made it an attractive candidate for further evaluation. More specifically, Healitide-GP1 had a hydrophobicity score of 0.44, a net charge of −2.58 and a QSAR score of 0.95, which were comparable to or better than other generated peptides. Compared to GP-WHP8, Healitide-GP1 had a slightly lower QSAR score, but its bioactivity and biocompatibility profiles were more promising. Therefore, we selected Healitide-GP1 as the main candidate for further experimental validation and characterization.

Therefore, Healitide-GP1 was selected as the lead candidate due to its optimal balance of QSAR score (0.95), hydrophobicity (0.44) and net charge (−2.58), which correlated with superior bioactivity in preliminary assays, as shown in Table 5.

Bioinformatics analysis

Based on the analysis, the novel WHP, GP-WHP1, with the sequence MFTMKKPLLLLFFLGTISLSLCEEE was identified as a promising candidate for wound healing. This peptide, generated by the genetic algorithm, showed desirable properties for wound healing and performed well in the experiments performed. The integrated interpretation of the peptide sequence alignment and the phylogenetic tree is shown in Figs. 4 and 5. The phylogenetic tree and sequence alignment provide complementary insights into the potential function and evolutionary relationships of the designed peptide compared to the 34 known or suspected WHPs. The phylogenetic tree shows that the designed peptide occupies a unique position and forms a distinct branch separate from the other peptides analyzed. This indicates that the designed peptide may have novel structural or functional properties. It is not closely related to known peptide families such as defensins (human defensin 5, human beta defensin 2 and 3), cathelicidin (LL-37, cathelicidin-DM) or histatin (human histatin 1). The closest branches are associated with chensinin-1b, indolicidin and myticin C. However, the considerable length of the branches separating the designed peptide from these neighbors indicates substantial sequence dissimilarity. The sequence alignment confirms the representation of the uniqueness of the designed peptide in the tree structure. The designed peptide exhibits a relatively low degree of sequence conservation with the other peptides, particularly at the N and C termini. Although there are some short stretches with similarities, the entire sequence, especially the central region (“LFFLGTISLSLCEEE”), is different. The unique composition of this central region, including a cluster of hydrophobic residues, supports the notion of a potentially novel mechanism of action. The presence of positively charged residues, as found in antibacterial peptides, was detected, suggesting possible interactions with bacterial membranes or immune components. The integration of the phylogenetic and sequence data strongly suggests that the designed peptide represents a novel peptide with a potentially unique role in wound healing. Its distinct evolutionary pathway is indicated by the tree and is consistent with its unique sequence features highlighted by the alignment.

The helical wheel projection of the designed peptide is shown in Fig. 6. The helical wheel projection displays the amino acid residues of the peptide arranged as they would appear in an alpha-helix in some section.

Based on I-TASSER modeling, the predicted secondary structure of healitide-GP1 revealed a mixture of alpha helices, beta strands and coils. This prediction suggests that the peptide adopts a helical structure in the middle region (residues 6–12), followed by a short beta-strand (residues 13–15) and a coil region at the C-terminus. This structural arrangement could contribute to the stability and functionality of the peptide. The predicted secondary structure of healitide-GP1, which comprises a mixture of alpha helices, beta strands and coils, may contribute to its stability and functionality by facilitating specific interactions with biological molecules such as membranes and receptors and creating a balance between rigidity and flexibility that is essential for its wound-healing and antimicrobial activities.

The predicted solvent accessibility profile shows that the peptide contains a mixture of buried and exposed residues. The N-terminus and helical region are relatively exposed, while the beta-strand and coil regions are more buried. The predicted B-factor profile suggests that the peptide has a moderate degree of flexibility, with the helical region being stiffer than the coil region. The top 10 threading templates used by I-TASSER show that the peptide has structural similarity to various proteins, including enzymes and receptors. The top 5 models predicted by I-TASSER have C-scores between − 1.05 and − 5, indicating varying levels of confidence in the predictions. Model 1 with a C-score of −1.05 is the most reliable prediction and may represent the most plausible structure for Healitide-GP1 (Fig. 6 right). The TM-align structural alignment program identified several proteins in the PDB library that have structural similarity to healitide-GP1. The top 10 proteins had TM scores between 0.701 and 0.727, indicating moderate to high structural similarity. These proteins may provide insights into the potential function and binding sites of Healitide-GP1. The COFACTOR and COACH servers predicted several potential functional sites on Healitide-GP1, including ligand binding sites, enzyme commission (EC) numbers, and Gene Ontology (GO) terms. The top 5 predicted ligand binding sites had C-scores ranging from 0.27 to 0.08, indicating varying levels of confidence in the predictions. The predicted EC numbers and GO terms suggest that healitide-GP1 may be involved in various biological processes, including metabolic pathways and cellular signaling.

Peptide design and synthesis

Healitide-GP1 was successfully synthesized with high purity (96.8%), which was confirmed by HPLC analysis. To confirm the successful synthesis and purity of Healitide-GP1, mass spectrometry analysis was performed. As shown in Fig. 7, the mass spectrometry chromatogram shows a dominant peak at 2703.2 m/z, which corresponds to the protonated molecular ion [M + H] + of healitide-GP1. This peak confirms the expected molecular weight of 2754.3 Da predicted from the peptide sequence. In addition, several other peaks representing different charge states and fragmentation ions were observed, further confirming the structure and purity of the synthesized peptide. The mass spectrometric data provide crucial evidence for the successful production of healitide-GP1 and ensure its suitability for subsequent functional testing.

The functional properties of Healitide-GP1 are summarized in Table 5, highlighting its high purity, cytocompatibility, accelerated wound closure, pro-angiogenic activity and strong antibacterial activity.

Cytotoxicity assay

The results of the MTT assay, as shown in Fig. 8, indicate that Healitide-GP1 does not exhibit significant cytotoxicity up to a concentration of 100 µg/mL in both HDF and HaCaT cells (p > 0.05 compared to control). Mild toxicity (~ 15) was observed at a concentration of 200 µg/mL, suggesting that concentrations below this threshold are safe for cell treatment. This finding confirms the cytocompatibility of Healitide-GP1 and supports its use in subsequent functional assays without fear of adverse effects on cell viability.

Wound healing assay

The wound healing activity of Healitide-GP1 was investigated using an in vitro scratch test. HDFs and HaCaT cells were treated with 50 µg/mL Healitide-GP1 for 24 h. The results showed that Healitide-GP1 significantly accelerated wound closure compared to the untreated group. In HDFs, 50 µg/mL Healitide-GP1 increased wound closure by 48% after 24 h, compared to 22% in the control group (p < 0.01). Similar effects were observed in HaCaT cells, with a closure rate of 52% compared to 25% in the control group (p < 0.01) (Fig. 9).

Antibacterial activity

Healitide-GP1 showed strong antimicrobial activity against both bacterial strains tested (Table 6). The MIC values were 12.5 µg/mL for S. aureus and 25.0 µg/mL for E. coli. The MBC values were 25.0 µg/mL for S. aureus and 50.0 µg/mL for E. coli, resulting in MBC/MIC ratios of 2.0 for both strains. All experiments were performed with n = 6 replicates. These ratios (≤ 4) classify Healitide-GP1 as bactericidal and not bacteriostatic, indicating that the peptide kills bacteria and not just inhibits their growth. This bactericidal activity is particularly advantageous for applications in wound healing, as it can effectively eliminate bacterial contamination and prevent biofilm formation. The different activity observed in Gram-positive and Gram-negative bacteria (2-fold higher MIC for E. coli) is consistent with the typical behavior of cationic antibacterial peptides. The increased activity against S. aureus can be attributed to the simpler cell wall structure of Gram-positive bacteria, which facilitates the interaction of the peptide with the cytoplasmic membrane. The slightly reduced activity against E. coli reflects the additional barrier provided by the outer membrane of Gram-negative bacteria, although the peptide remains highly effective. These antibacterial results correlate strongly with the predicted cationic and amphipathic nature of Healitide-GP1 as determined by computational analysis. The presence of positively charged residues (lysine and arginine) enabled electrostatic interaction with negatively charged bacterial membranes, while the amphipathic structure facilitated membrane insertion and disruption. This mechanism of action supports the dual function of Healitide-GP1 in both wound healing and microbial control, making it an ideal candidate for the treatment of infected wounds. The antibacterial efficacy of Healitide-GP1 at concentrations well below those used for wound healing (50 µg/mL) suggests that the peptide may simultaneously promote tissue repair and prevent bacterial infection. This dual activity is particularly valuable in the clinical treatment of wounds, where bacterial contamination often impedes the healing process and can lead to the formation of chronic wounds.

WHPs are promising therapeutic targets for accelerating tissue repair and regeneration¹⁷. This study aimed to classify WHPs and non-WHPs using machine learning and generate novel WHPs using a genetic algorithm. By leveraging computational tools, we identified features critical to WHP activity and generated potential WHPs, some of which had no significant similarity to known sequences, indicating their novelty.

Model performance

To classify WHPs and non-WHPs, several machine learning models were trained and evaluated. Among the models tested, the Light Gradient Boosting Machine (LightGBM) emerged as the best-performing model with an accuracy of 0.89, AUC-ROC of 0.90, and MCC of 0.81. These metrics indicate that the model can classify peptides with high precision and recall, making it suitable for WHP prediction.

Accuracy is a critical metric for evaluating machine learning models because it measures the proportion of correct predictions¹⁸. Our LightGBM model correctly classified 89% of the test dataset, which is a significant improvement compared to other models tested, such as the Gradient Boosting Classifier (accuracy = 0.85) and Random Forest (accuracy = 0.86). Additionally, the AUC-ROC value of 0.9 underscores the model’s ability to distinguish between WHPs and non-WHPs across various classification thresholds effectively. The MCC value of 0.81 further supports the robustness of the model because the MCC is a balanced metric that considers true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN), making it less sensitive to class imbalance.

The LightGBM model demonstrated a minor bias toward predicting non-WHPs as WHPs. While this bias is acceptable, it provides higher confidence in peptides predicted as WHPs because false positives (non-WHPs predicted as WHPs) are minimized. This conservative approach ensures that WHP peptides are more likely to exhibit true wound-healing activity.

Key features of the WHP classification

The model identified several key features that contributed significantly to the classification of WHPs and non-WHPs. The amino acid composition (ACC), DPC (DPC), hydrophobicity, and net charge were the most important.

1. (1)

   Cysteine (C): Cysteine composition was the most negatively correlated feature with WHP activity, with a correlation coefficient of -0.81. The low abundance of this peptide in WHPs is consistent with findings from other studies, which suggest that cysteine-rich peptides are more commonly associated with antibacterial activity than wound healing^19,20.

2. (2)

   Lysine (K): Lysine composition was the most positively correlated feature, with a correlation coefficient of 0.86. Lysine’s positive charge enhances peptide interactions with negatively charged cell membranes, facilitating tissue repair and antibacterial activity²¹.

3. (3)

   DPC (GK, LL): The dipeptide GK (Glycine-Lysine) was identified as a significant feature, reflecting its importance in WHPs, which are known for their tissue-repair properties²². Similarly, LL (Leucine-Leucine) was another critical dipeptide, as leucine is a key amino acid in amphipathic peptides that penetrate cell membranes to promote healing²³.

4. (4)

   Hydrophobicity: Hydrophobicity is a major contributor to the WHP classification. Hydrophobic residues enable peptides to interact with lipid membranes, which is a crucial step in cellular repair and signaling.

5. (5)

   Net Charge: Most WHPs are cationic, allowing them to bind to negatively charged cell membranes, which is essential for wound closure and regeneration²⁴. The net charge on the WHPs is a distinguishing feature of the proposed model.

These findings align with those of previous studies that have emphasized the importance of specific amino acids and physicochemical properties in determining peptide bioactivity^25,26,27.

Generation of novel WHPs

Using a genetic algorithm (GA), we generated novel WHPs (GP-WHPs) in two iterations. In the first iteration, no sequence length restrictions were applied, resulting in 30 peptides, of which 5 had no significant similarity to known sequences in the NCBI database. In the second iteration, the sequence length was restricted to ≤ 100 amino acids, yielding 12 peptides, including 4 sequences with no functional annotation.

Novelty and functional insights

Among the generated peptides:

• GP-WHP1 was identified as a novel peptide with no significant similarity in the database. It achieved a high QSAR score of 0.99 for wound-healing activity and exhibited favorable physicochemical properties, including a hydrophobicity value of 0.44 and a lysine composition of 10.9%. These properties suggest a strong potential for cellular interactions and membrane lysis, which are critical for wound healing²⁸.

• GP-WHP2 is another promising AMP candidate, with a QSAR score of 0.89. This peptide has a high lysine content and DPC (GK), further supporting its potential for wound-healing applications.

Functional annotation of homologous sequences

Some generated peptides were significantly similar to the sequences in the NCBI database:

• TRZ08009.1: A hypothetical protein from Zosterops borbonicus with 69% sequence identity with GP-WHP3. Despite its unknown function, it was predicted as a WHP by AMPDiscover with a QSAR score of 0.89.

• CAA0409135: An unnamed protein from Arabidopsis thaliana with 81.9% sequence similarity to GP-WHP4. This peptide was predicted as AMP with QSAR scores of 0.78 (AMP) and 0.8 (antibacterial).

Unique peptides without similarity

The GP-WHP5 and GP-WHP6 peptides were not significantly similar to any known sequences. These peptides represent truly novel candidates for wound-healing applications. GP-WHP5, in particular, had a high QSAR score of 0.72 for wound-healing activity, whereas GP-WHP6 achieved a perfect QSAR score of 1.0, indicating exceptional potential.

Comparison with previous studies

Our LightGBM model achieved an AUC-ROC of 0.90, outperforming previous methods, such as the gradient boosting machine model, which achieved an accuracy of 0.63 on antibacterial peptide datasets. The results demonstrate the effectiveness of the proposed feature selection and model optimization approach for WHP classification. In addition, the inclusion of physicochemical properties and DPC enhanced the model performance, as evidenced by the high MCC and AUC-ROC values.

Implications of the WHP design

This study highlights the potential of machine learning and genetic algorithms in peptide design. By identifying key features such as lysine composition, hydrophobicity, and specific dipeptides, we can guide the rational design of WHPs. The generation of novel peptides with no significant similarity to existing sequences offers opportunities for further experimental validation and therapeutic development.

Potential of a novel mechanism

The combination of evolutionary divergence and sequence uniqueness implies that the designed peptide may interact with different targets or employ a different mechanism of action compared with other WHPs. The hydrophobic patch in the central region is a key feature driving this distinct functionality²⁹.

Possible antibacterial/immunomodulatory activity

Although the designed peptide does not cluster with known antibacterial families, the presence of positively charged residues, coupled with some limited sequence similarity to peptides with potential antibacterial properties, suggests that it might possess some degree of antibacterial or immunomodulatory activity³⁰.

Relationship with neighboring peptides

The phylogenetic proximity to Chensinin-1b, Indolicidin, and myticin C, although distant, warrants further investigation. It is possible that the designed peptide shares some functional or structural features with these peptides, despite the overall sequence divergence. Based on the sequence, it appears that the designed peptide has similar characteristics to antibacterial peptides^31,32. The designed peptide was named Healitide-GP1.

The arrangement in the helical wheel reveals the distribution of the hydrophobic and hydrophilic residues around the helix’s circumference. The diagram clearly shows the segregation of the hydrophobic (nonpolar) and hydrophilic (polar) amino acids. One side of the helix is predominantly composed of hydrophobic residues (L, F, M, I), whereas the other side contains more hydrophilic residues (K, T, E, S, C). This amphipathic nature is critical for the peptide’s potential function in wound healing^33,34. The hydrophobic face is rich in Leucine (L), Phenylalanine (F), Methionine (M), and Isoleucine (I)) is likely to interact with the hydrophobic core of the cell membrane or other hydrophobic components in the wound bed³⁵. This interaction may facilitate cell penetration or interaction with the extracellular matrix. The hydrophilic face (containing Lysine (K), Threonine (T), Glutamic acid (E), Serine (S), and Cysteine (C)) is likely to interact with the aqueous environment of the wound or with polar molecules involved in the healing process^36,37. This face could mediate interactions with growth factors, cytokines, and other proteins crucial for tissue repair. This arrangement suggests that the peptide interacts with cell membranes to facilitate cellular processes^38,39. The presence of both hydrophobic and hydrophilic amino acids indicates that the peptide is likely to have multiple interaction points, potentially enhancing its activity. The presence of a single cysteine residue is also noteworthy⁴⁰. This potentially allows for the dimerization or oligomerization of the peptide through disulfide bond formation, thereby influencing the overall structure and biological activity⁴¹. The amphipathic nature of the peptide is a favorable characteristic of wound-healing agents. The hydrophobic face could promote interactions with the cell membrane or hydrophobic components of the extracellular matrix, whereas the hydrophilic face could contribute to interactions with growth factors and cytokines. These interactions can stimulate cellular processes involved in wound healing, such as cell migration, proliferation, and differentiation³. In summary, the helical wheel projection suggests that the designed peptide has a promising amphipathic structure, which could contribute to its potential as a wound-healing agent. However, experimental validation is crucial for confirming its efficacy and safety.

Integration of functional validation with computational design

The successful synthesis and functional validation of Healitide-GP1 confirm the effectiveness of our machine learning-based design pipeline. The combination of feature importance analysis (e.g. lysine content, hydrophobicity and specific dipeptides) with sequence optimization guided by genetic algorithms resulted in a novel peptide with robust wound healing properties. Evaluation of the cytocompatibility of Healitide-GP1 revealed an excellent safety profile, which is crucial for therapeutic applications. The peptide exhibited minimal cytotoxicity of up to 100 µg/ml in both HDF and HaCaT cells, with cell viability above 95%. This result is particularly important as it opens a broad therapeutic window for clinical applications. The concentration used in our wound healing assays (50 µg/mL) showed optimal efficacy at > 97% cell viability, suggesting that therapeutic benefits can be achieved without compromising cell health. The mild cytotoxicity observed at 200 µg/mL (~ 15% reduction in viability) is consistent with the general behavior of cationic antibacterial peptides, which can interact with mammalian cell membranes at high concentrations due to their amphipathic nature^42,43. However, the fact that viability remained above 84% even at this high concentration suggests that Healitide-GP1 has favorable selectivity for bacterial membranes over mammalian cells. This selectivity can be attributed to differences in membrane composition, with bacterial membranes containing a higher proportion of negatively charged phospholipids that preferentially interact with cationic peptides⁴⁴. The similar cytotoxicity profiles observed in both fibroblasts and keratinocytes suggest that Healitide-GP1 does not exhibit cell type-specific toxicity, which is advantageous for applications in wound healing where multiple cell types are involved in the repair process. This broad compatibility suggests that the peptide can be safely applied to different wound types without having to worry about the different cellular responses. The strong antibacterial activity of Healitide-GP1 against both Gram-positive (S. aureus) and Gram-negative (E. coli) bacteria represents a significant advantage for applications in wound healing. The MIC values of 12.5 µg/ml and 25.0 µg/ml are comparable to or better than many conventional antibacterial peptides reported in the literature^45,46.

To contextualize the therapeutic potential of Healitide-GP1, we conducted a comprehensive comparison with established wound-healing AMPs documented in recent literature (Table 7).

Healitide-GP1 showed several clear advantages over comparable peptides. While most wound-healing AMPs carry high positive charges (+ 4 to + 11), the unique anionic character of healitide-GP1 (−2.58) represents a novel mechanism that potentially bypasses the resistance pathways associated with cationic peptides⁵⁵. This structural difference is particularly significant as resistance to cationic AMPs such as LL-37 has been demonstrated in clinical isolates of S. aureus through modifications of membrane charge⁵⁶. The antimicrobial efficacy of Healitide-GP1 (MIC: 12.5 µg/mL against S. aureus, 25 µg/mL against E. coli) equaled or exceeded that of clinically evaluated peptides. These values are comparable to those of pexiganan (MSI-78), which has reached Phase III clinical trials, and exceed LL-37, the most extensively studied human AMP. Unlike pexiganan, which was denied FDA approval due in part to cytotoxicity concerns⁵⁷, Healitide-GP1 maintained > 95% cell viability at concentrations up to 100 µg/ml, which represents a 10-fold improvement in the margin of safety compared to LL-37, which exhibits cytotoxicity above 10 µg/ml. In terms of wound healing efficiency, Healitide-GP1 achieved 48–52% wound closure at 24 h, outperforming most established peptides such as LL-37 (35–45%), pexiganan (40–45%) and hBD-3 (40–50%)^51,58. This improved wound closure rate combined with the peptide’s favorable safety profile results in a therapeutic index (IC50/MIC) of > 8.0, which is 2to 4 times higher than typical wound-healing AMPs reported in the literature⁵⁸. The comparison also showed that many AMPs in clinical development face problems such as salt sensitivity (hBD-2), hemolytic activity or rapid proteolytic degradation, whereas Healitide-GP1 showed stability under physiological conditions without detectable hemolysis at concentrations of up to 200 µg/ml. These properties address critical limitations that have hindered the clinical implementation of AMPs in the past^59,60. In addition, the dual functionality of Healitide-GP1 – combining potent antimicrobial activity with healing-promoting properties through enhanced cell migration and proliferation- is an advantage over peptides that exhibit only one of these activities. This is particularly important for the treatment of chronic wounds, where both infection control and tissue regeneration are crucial for successful healing^61,62.

The 2-fold difference in activity between the two bacterial strains is typical of cationic antibacterial peptides and reflects the structural differences in bacterial cell walls. The bactericidal activity of Healitide-GP1, as evidenced by an MBC/MIC ratio of 2.0, is particularly valuable for applications in wound healing. In contrast to bacteriostatic agents, which merely inhibit bacterial growth, bactericidal peptides actively eliminate pathogens, reduce the risk of resistance development and prevent the formation of biofilms^63,64. This is particularly important in the treatment of chronic wounds, where bacterial biofilms are a major obstacle to healing.

Based on the amphipathic structure and anionic nature of Healitide-GP1, we hypothesize that the antibacterial mechanism is similar to other anionic AMPs (AAMPs). AAMPs have a different bioactive mechanism compared to cationic AMPs. AAMPs can interact with the negatively charged microbial membranes via metal ions such as zinc or calcium. These ions act as a bridge between AAMPs and microbial membranes and facilitate binding and destruction. These anionic peptides therefore also have membrane and intracellular targets. These peptides disrupt the membrane by forming pores (like the toroidal pore model). They also translocate inside the cell to inhibit targets such as ribosomes or, in some cases, to form intracellular amyloid structures^4965,66. Similar to cationic AMPs, AAMPs have hydrophobic and hydrophilic domains that allow interaction with lipid bilayers, but through different mechanisms with microbial cells⁶⁷. The amphipathic structure of the peptide with a hydrophobic and a hydrophilic side enables it to interact with and destroy bacterial membranes⁶⁸. Studies have shown that AMPs can interact with bacterial membranes through electrostatic interactions between the positively charged amino acids of the peptide and the negatively charged phospholipids of the membrane^49,69. This interaction can lead to a disruption of the membrane, either by the formation of pores or by the complete destruction of the membrane structure70. These proposed mechanism can be considered for Healitide-GP1. The mechanism of action of Healitide-GP1 is probably based on a similar process in which both membrane and intracellular targets, including vital enzymes or metabolic processes, are disrupted. This disruption can result in the loss of cellular contents, including ions, nutrients and genetic material, ultimately leading to bacterial death⁷¹. The bactericidal activity of the peptide, as evidenced by its low peptide interacts with the bacterial membrane and disrupts its structure via cationic ions MBC/MIC ratio, suggests that it is capable of effectively killing bacterial cells rather than just inhibiting their growth. This is consistent with the notion that healitide-GP1 is able to disrupt bacterial membranes and cause cell death. A study on the antimicrobial peptide LL-37 found that it interacts with bacterial membranes through electrostatic interactions and disrupts the membrane structure, leading to bacterial death⁷². Another study on the antimicrobial peptide melittin found that it forms pores in bacterial membranes, leading to loss of cell contents and bacterial death⁷³. In a review of antimicrobial peptides, it was found that they can interact with bacterial membranes through various mechanisms, including electrostatic interactions, hydrophobic interactions and membrane disruption⁴³. Overall, the antibacterial mechanism of Healitide-GP1 is likely due to membrane disruption, which is a common mechanism of action for cationic antimicrobial peptides. The amphipathic structure of the peptide and its bactericidal activity support this interpretation, and relevant references provide further evidence in favor of this mechanism.

The combination of wound healing and antibacterial activity in a single peptide represents a significant advance in wound care therapy. Conventional wound treatments often require separate antibacterial and healing agents, which can lead to complex treatment regimens and potential drug interactions. The dual functionality of Healitide-GP1 addresses both aspects of wound care simultaneously, which could simplify treatment protocols and improve patient compliance. The therapeutic concentrations required for antibacterial activity (12.5–25.0 µg/ml) are well below those used for wound healing (50 µg/ml), ensuring that both activities can be achieved simultaneously without compromising safety. This overlap of therapeutic ranges is particularly beneficial for the treatment of infected wounds, where bacterial contamination often impedes the natural healing process. The selectivity indices calculated for Healitide-GP1 (7.6 for S. aureus and 3.8 for E. coli) indicate a favorable selectivity for bacterial cells over mammalian cells. These values exceed the minimum threshold (SI > 2) normally required for therapeutic peptides, supporting the clinical potential of Healitide-GP1^74,75. Current WHPs and antibacterial agents often have limitations in terms of efficacy or safety. Many antibacterial peptides exhibit excellent antibacterial activity but lack wound-healing properties, while growth factors and healing-promoting peptides may not be effective against bacterial contamination^76,77. The dual functionality of Healitide-GP1 makes it a superior alternative to existing monofunctional treatments. In addition, the novel sequence of healitide-GP1, which bears no significant similarity to known peptides, suggests a potentially unique mechanism of action that could overcome the limitations associated with existing therapies. The machine learning-based design approach used to develop healitide-GP1 represents a paradigm shift from traditional empirical methods and enables the rational design of peptides with optimized properties.

Comparison with existing WHPs

Healitide-GP1 is a novel WHP that exhibits dual functionality, namely both antibacterial and wound-healing properties. To better understand its potential as a therapeutic agent, it is important to compare its efficacy and toxicity profile with existing WHPs. One of the best known WHPs is LL-37, a cathelicidin-derived peptide that has been shown to exhibit broad-spectrum antimicrobial activity and promote wound healing⁷⁸. Studies have shown that LL-37 can accelerate wound closure, reduce bacterial load and promote tissue repair^72,79. However, LL-37 has also been shown to have cytotoxic effects on mammalian cells at high concentrations⁸⁰. In comparison, Healitide-GP1 has been shown to have a similar wound closure rate to LL-37, with a 48% increase in wound closure after 24 h. However, Healitide-GP1 has a lower MIC than LL-37, with an MIC of 12.5 µg/mL against Staphylococcus aureus. In addition, healitide-GP1 has been shown to have a lower toxicity profile than LL-37, with no significant cytotoxicity observed at concentrations up to 100 µg/mL⁷⁹. Cathelicidin-NV, a peptide from Nanorana ventripunctata, has been shown to promote wound re-epithelialization and accelerate healing⁸¹. This peptide also exhibits antioxidant activity, which is beneficial for wound repair⁸¹. Healitide-GP1, on the other hand, has been shown to have a broader spectrum of antimicrobial activity, with activity against both Gram-positive and Gram-negative bacteria. B-2Ta, a peptide from Pelophylax kl. esculentus, has been shown to have both antimicrobial and wound-healing properties⁸². Healitide-GP1 has a similar antimicrobial activity profile to B-2Ta, but with a lower toxicity profile, as no significant cytotoxicity was observed at concentrations of up to 100 µg/mL. Tet213-CN improves bacterial phagocytosis and promotes M1 macrophage polarization^83,84. Due to observed potent antibacterial activity, Healitide-GP1 can be have a similar ability to promote M1 macrophage polarization, which is essential for effective antimicrobial activity. Mast-MO compromises bacterial cell membrane integrity and stimulates cytokine production⁸⁵. The mechanism of antibacterial activity for Healitide-GP1 may be similar to Mast-MO peptide. Andersonin-W1 (AW1) directly kills pathogens and prevents biofilm formation⁸⁶. Healitide-GP1 may be able to prevent biofilm formation, which is essential for effective wound healing process. Antimicrobial peptide derived from insulin-like growth factor-binding protein 5 (AMP-IBP5) exhibits broad-spectrum antibacterial effects against Gram-positive and Gram-negative bacteria^87,88. Healitide-GP1 has a similar antibacterial activity profile to AMP-IBP5, with activity against both Gram-positive and Gram-negative bacteria. Defensins are another class of WHPs that have been shown to have antibacterial and wound-healing properties⁸⁹. For example, human defensin 1 (HD6) has been shown to have a broad spectrum of antibacterial activity and promotes wound healing⁹⁰. However, defensins can also have cytotoxic effects on mammalian cells⁹¹. In comparison, Healitide-GP1 has been shown to have a similar antibacterial activity profile to defensins, with a broad spectrum of activity against Gram-positive and Gram-negative bacteria. However, Healitide-GP1 has a lower toxicity profile than defensins, as no significant cytotoxicity was observed at concentrations of up to 100 µg/ml. Overall, Healitide-GP1 appears to possess a unique combination of antimicrobial and wound-healing properties with a low toxicity profile, making it a promising therapeutic candidate for the treatment of infected wounds.

1. Wilkinson, H. N. & Hardman, M. J. Wound healing: Cellular mechanisms and pathological outcomes. Open. Biology. 10 (9), 200223 (2020).

   PubMed  PubMed Central  Google Scholar 

2. Kolimi, P., Narala, S., Nyavanandi, D., Youssef, A. A. A. & Dudhipala, N. Innovative treatment strategies to accelerate wound healing: Trajectory and recent advancements. Cells 11 (15), 2439 (2022).

   PubMed  PubMed Central  Google Scholar 

3. Md Fadilah, N. I. et al. Discovery of bioactive peptides as therapeutic agents for skin wound repair. J. Tissue Eng. 15, 20417314241280359 (2024).

   PubMed  PubMed Central  Google Scholar 

4. Shariatzadeh, F. J., Currie, S., Logsetty, S., Spiwak, R. & Liu, S. Enhancing wound healing and minimizing scarring: A comprehensive review of nanofiber technology in wound dressings. Prog. Mater. Sci. 147, 101350 (2024).

5. Khan, N. U. et al. Obstructed vein delivery of ceftriaxone via poly(vinyl-pyrrolidone)-iodine-chitosan nanofibers for the management of diabetic foot infections and burn wounds. Int. J. Biol. Macromol. 277, 134166 (2024).

   PubMed  Google Scholar 

6. Ni, J. et al. Epigallocatechin gallate and Vancomycin loaded poly(vinyl)-pyrrolidone-gelatine nanofibers, conceivable curative approach for wound healing. Colloids Surf., B. 249, 114506 (2025).

   Google Scholar 

7. Khan, N. U. et al. α-Lactalbumin based scaffolds for infected wound healing and tissue regeneration. Int. J. Pharm. 663, 124578 (2024).

   PubMed  Google Scholar 

8. Adnan, S. B., Maarof, M., Fauzi, M. B. & Md Fadilah, N. I. Antimicrobial peptides in wound healing and skin regeneration: Dual roles in immunity and microbial defense. Int. J. Mol. Sci. 26 (13), 5920 (2025).

   PubMed  PubMed Central  Google Scholar 

9. Kanaujia, K. A., Mishra, N., Rajinikanth, P. S. & Saraf, S. A. Antimicrobial peptides as antimicrobials for wound care management: A comprehensive review. J. Drug Deliv. Sci. Technol. 95, 105570 (2024).

   Google Scholar 

10. Brizuela, C. A., Liu, G., Stokes, J. M. & de la Fuente-Nunez, C. AI methods for antimicrobial peptides: Progress and challenges. Microb. Biotechnol. 18 (1), e70072 (2025).

    PubMed  PubMed Central  Google Scholar 

11. Yang, R. et al. Advances in antimicrobial peptides: From mechanistic insights to chemical modifications. Biotechnol. Adv. 81, 108570 (2025).

    PubMed  Google Scholar 

12. Chen, Z., Wang, R., Guo, J. & Wang, X. The role and future prospects of artificial intelligence algorithms in peptide drug development. Biomed. Pharmacother. 175, 116709 (2024).

    PubMed  Google Scholar 

13. Yang, S. & Kar, S. Application of artificial intelligence and machine learning in early detection of adverse drug reactions (ADRs) and drug-induced toxicity. Artif. Intell. Chem. 1 (2), 100011 (2023).

    Google Scholar 

14. Sun, S. Progress in the identification and design of novel antimicrobial peptides against pathogenic microorganisms. Probiotics Antimicrob. Proteins. 17 (2), 918–936 (2025).

    PubMed  Google Scholar 

15. Lata, M. et al. Evolutionary and in Silico guided development of novel peptide analogues for antibacterial activity against ESKAPE pathogens. Curr. Res. Microb. Sci. 4, 100183 (2023).

    PubMed  PubMed Central  Google Scholar 

16. Hashemi, S., Vosough, P., Taghizadeh, S. & Savardashtaki, A. Therapeutic peptide development revolutionized: Harnessing the power of artificial intelligence for drug discovery. Heliyon. 10, (2024).

17. Lee, Y-C-J., Javdan, B., Cowan, A. & Smith, K. More than skin deep: Cyclic peptides as wound healing and cytoprotective compounds. Front. Cell. Dev. Biol. 11, 1195600 (2023).

    PubMed  PubMed Central  Google Scholar 

18. Deng, Y., Eden, M. R. & Cremaschi, S. Metrics for evaluating machine learning models prediction accuracy and uncertainty. In Computer Aided Chemical Engineering vol. 52, pp. 1325–30 (Elsevier, 2023).

19. Dini, I., De Biasi, M-G. & Mancusi, A. An overview of the potentialities of antimicrobial peptides derived from natural sources. Antibiotics 11 (11), 1483 (2022).

    PubMed  PubMed Central  Google Scholar 

20. Aquib, M., Sharma, T., Giri, V. P. & Rao, A. Peptides in chronic wound healing: Fighting infections and facilitating diabetic wound management. J. Drug Deliv. Sci. Technol. 105, 106599 (2025).

21. Zheng, M. et al. Poly (α-l-lysine)-based nanomaterials for versatile biomedical applications: Current advances and perspectives. Bioactive Mater. 6 (7), 1878–1909 (2021).

    Google Scholar 

22. Yin, S., Wang, Y. & Yang, X. Amphibian-derived wound healing peptides: Chemical molecular treasure trove for skin wound treatment. Front. Pharmacol. 14, 1120228 (2023).

    PubMed  PubMed Central  Google Scholar 

23. Rehman, S. U., Ali, R., Zhang, H., Zafar, M. H. & Wang, M. Research progress in the role and mechanism of leucine in regulating animal growth and development. Front. Physiol. 14, 1252089 (2023).

    PubMed  PubMed Central  Google Scholar 

24. Zhang, Q-Y. et al. Antimicrobial peptides: Mechanism of action, activity and clinical potential. Military Med. Res. 8, 1–25 (2021).

    ADS  Google Scholar 

25. Akbarian, M., Khani, A., Eghbalpour, S. & Uversky, V. N. Bioactive peptides: Synthesis, sources, applications, and proposed mechanisms of action. Int. J. Mol. Sci. 23(3), 1445 (2022).

26. Dong, Y. & Dai, Z. Physicochemical properties, biological activities, and peptidome profile of grass carp swim bladder collagen hydrolysates. LWT 189, 115544 (2023).

    Google Scholar 

27. Akbarian, M., Khani, A., Eghbalpour, S. & Uversky, V. N. Bioactive peptides: Synthesis, sources, applications, and proposed mechanisms of action. Int. J. Mol. Sci. 23 (3), 1445 (2022).

    PubMed  PubMed Central  Google Scholar 

28. Hosty, L., Heatherington, T., Quondamatteo, F. & Browne, S. Extracellular matrix-inspired biomaterials for wound healing. Mol. Biol. Rep. 51 (1), 830 (2024).

    PubMed  PubMed Central  Google Scholar 

29. Lang, S. et al. Signal peptide features determining the substrate specificities of targeting and translocation components in human ER protein import. Front. Physiol. 11, 833540 (2022).

30. Talapko, J. et al. Antimicrobial peptides—mechanisms of action, antimicrobial effects and clinical applications. Antibiotics 11 (10), 1417 (2022).

    PubMed  PubMed Central  Google Scholar 

31. Huan, Y., Kong, Q., Mou, H. & Yi, H. Antimicrobial peptides: Classification, design, application and research progress in multiple fields. Front. Microbiol. 11, 582779 (2020).

    PubMed  PubMed Central  Google Scholar 

32. Dong, W., Mao, X., Guan, Y., Kang, Y. & Shang, D. Antimicrobial and anti-inflammatory activities of three chensinin-1 peptides containing mutation of Glycine and histidine residues. Sci. Rep. 7 (1), 40228 (2017).

    ADS  PubMed  PubMed Central  Google Scholar 

33. Bui Thi Phuong, H., Doan Ngan, H., Le Huy, B., Vu Dinh, H. & Luong Xuan, H. The amphipathic design in helical antimicrobial peptides. ChemMedChem 19 (7), e202300480 (2024).

    PubMed  Google Scholar 

34. Zhou, J. et al. An injectable, natural peptide hydrogel with potent antimicrobial activity and excellent wound healing-promoting effects. Nano Today. 49, 101801 (2023).

    Google Scholar 

35. Al Mughram, M. H., Catalano, C., Herrington, N. B., Safo, M. K. & Kellogg, G. E. 3D interaction homology: The hydrophobic residues alanine, isoleucine, leucine, proline and valine play different structural roles in soluble and membrane proteins. Front. Mol. Biosci. 10, 1116868 (2023).

    PubMed  PubMed Central  Google Scholar 

36. Pikuła, M. et al. Cystatin C peptidomimetic derivative with antimicrobial properties as a potential compound against wound infections. Bioorg. Med. Chem. 25 (4), 1431–1439 (2017).

    ADS  PubMed  Google Scholar 

37. Trejos, M., Aristizabal, Y., Aragón-Muriel, A., Oñate-Garzón, J. & Liscano, Y. Characterization and classification in Silico of peptides with dual activity (antimicrobial and wound healing). Int. J. Mol. Sci. 24 (17), 13091 (2023).

    PubMed  PubMed Central  Google Scholar 

38. Nurkesh, A., Jaguparov, A., Jimi, S. & Saparov, A. Recent advances in the controlled release of growth factors and cytokines for improving cutaneous wound healing. Front. Cell. Dev. Biol. 8-2020, 638 (2020).

39. Gonçalves, S. & Santos, N. C. Membrane-Peptide interactions: From basics to current applications 2.0. Int. J. Mol. Sci. 24(8), 7202 (2023).

40. Bhargava, K. & Feix, J. B. Membrane binding, structure, and localization of cecropin-mellitin hybrid peptides: A site-directed spin-labeling study. Biophys. J. 86 (1), 329–336 (2004).

    ADS  PubMed  PubMed Central  Google Scholar 

41. Lorenzón, E. et al. Effect of dimerization on the mechanism of action of aurein 1.2. Biochim. Et Biophys. Acta (BBA)-Biomembr. 1858 (6), 1129–1138 (2016).

    Google Scholar 

42. He, S. & Deber, C. M. Interaction of designed cationic antimicrobial peptides with the outer membrane of gram-negative bacteria. Sci. Rep. 14 (1), 1894 (2024).

    ADS  PubMed  PubMed Central  Google Scholar 

43. Ma, X. et al. A review of antimicrobial peptides: Structure, mechanism of action, and molecular optimization strategies. Fermentation 10 (11), 540 (2024).

    Google Scholar 

44. Zare-Zardini, H. et al. From defense to offense: Antimicrobial peptides as promising therapeutics for cancer. Front. Oncol. 14, 1463088 (2024).

    PubMed  PubMed Central  Google Scholar 

45. Rima, M. et al. Antimicrobial peptides: A potent alternative to antibiotics. Antibiot. (Basel) 10(9), 1095 (2021).

46. Lohan, S. et al. Broad-spectrum activity of membranolytic cationic macrocyclic peptides against multi-drug resistant bacteria and fungi. Eur. J. Pharm. Sci. 197, 106776 (2024).

    PubMed  PubMed Central  Google Scholar 

47. Hancock, R. E. & Scott, M. G. The role of antimicrobial peptides in animal defenses. Proc. Natl. Acad. Sci. U.S.A. 97 (16), 8856–8861 (2000).

    ADS  PubMed  PubMed Central  Google Scholar 

48. Ahmad, A. et al. Design and characterization of short antimicrobial peptides using leucine zipper templates with selectivity towards microorganisms. Amino Acids. 46 (11), 2531–2543 (2014).

    PubMed  Google Scholar 

49. Zhang, Q. Y. et al. Antimicrobial peptides: Mechanism of action, activity and clinical potential. Mil Med. Res. 8 (1), 48 (2021).

    ADS  PubMed  PubMed Central  Google Scholar 

50. Gera, S., Kankuri, E. & Kogermann, K. Antimicrobial peptides – Unleashing their therapeutic potential using nanotechnology. Pharmacol. Ther. 232, 107990 (2022).

    PubMed  Google Scholar 

51. Yin, J. & Yu, F. S. LL-37 via EGFR transactivation to promote high glucose-attenuated epithelial wound healing in organ-cultured Corneas. Investig. Ophthalmol. Vis. Sci. 51 (4), 1891–1897 (2010).

    Google Scholar 

52. Mangoni, M. L., McDermott, A. M. & Zasloff, M. Antimicrobial peptides and wound healing: Biological and therapeutic considerations. Exp. Dermatol. 25 (3), 167–173 (2016).

    PubMed  PubMed Central  Google Scholar 

53. Ramos, R. et al. Wound healing activity of the human antimicrobial peptide LL37. Peptides 32 (7), 1469–1476 (2011).

    PubMed  Google Scholar 

54. Gibson, A. L. et al. Nonviral human beta defensin-3 expression in a bioengineered human skin tissue: A therapeutic alternative for infected wounds. Wound Repair. Regener. Off. Public. Wound Healing Soc. [and] Eur. Tissue Repair. Soc. 20 (3), 414–424 (2012).

    Google Scholar 

55. Ahmad, A., Khan, J. M. & Bandy, A. A systematic review of the design and applications of antimicrobial peptides in wound healing. Cureus 16 (4), e58178 (2024).

    PubMed  PubMed Central  Google Scholar 

56. Rai, A. et al. Antimicrobial Peptide-Tether dressing able to enhance wound healing by tissue contact. ACS Appl. Mater. Interfaces. 14 (21), 24213–24228 (2022).

    PubMed  Google Scholar 

57. Kumar, P., Kizhakkedathu, J. N. & Straus, S. K. Antimicrobial peptides: Diversity, mechanism of action and strategies to improve the activity and biocompatibility in vivo. Biomolecules 8(1), 4 (2018).

58. Chen, Z., Meng, L., Sethi, G., Wang, J. & Li, B. Peptide-driven approaches in advanced wound healing materials. Drug Discov Today . 30, 104440 (2025).

59. Greco, I. et al. Correlation between hemolytic activity, cytotoxicity and systemic in vivo toxicity of synthetic antimicrobial peptides. Sci. Rep. 10 (1), 13206 (2020).

    ADS  PubMed  PubMed Central  Google Scholar 

60. Starr, C. G. & Wimley, W. C. Antimicrobial peptides are degraded by the cytosolic proteases of human erythrocytes. Biochim. Et Biophys. Acta Biomembr. 1859 (12), 2319–2326 (2017).

    Google Scholar 

61. Eming, S. A., Martin, P. & Tomic-Canic, M. Wound repair and regeneration: Mechanisms, signaling, and translation. Sci. Transl. Med. 6 (265), 265sr6 (2014).

    PubMed  PubMed Central  Google Scholar 

62. Falcone, M. et al. Challenges in the management of chronic wound infections. J. Global Antimicrob. Resist. 26, 140–147 (2021).

    Google Scholar 

63. Fjell, C. D., Hiss, J. A., Hancock, R. E. & Schneider, G. Designing antimicrobial peptides: Form follows function. Nat. Rev. Drug Discovery. 11 (1), 37–51 (2012).

    Google Scholar 

64. Xuan, J. et al. Antimicrobial peptides for combating drug-resistant bacterial infections. Drug Resist. Updates. 68, 100954 (2023).

    Google Scholar 

65. Harris, F., Dennison, S. R. & Phoenix, D. A. Anionic antimicrobial peptides from eukaryotic organisms. Curr. Protein Pept. Sci. 10 (6), 585–606 (2009).

    PubMed  Google Scholar 

66. Arun, B., Rejeesh, E. & Rani, N. M. Future Perspective of Peptide Antibiotic Market 311–320 (Elsevier, 2023).

    Google Scholar 

67. Zhang, R., Xu, L. & Dong, C. Antimicrobial peptides: An overview of their structure, function and mechanism of action. Protein Pept. Lett. 29 (8), 641–650 (2022).

    PubMed  Google Scholar 

68. Stephani, J. C., Gerhards, L., Khairalla, B., Solov’yov, I. A. & Brand, I. How do antimicrobial peptides interact with the outer membrane of Gram-Negative bacteria? Role of lipopolysaccharides in peptide binding, anchoring, and penetration. ACS Infect. Dis. 10 (2), 763–778 (2024).

    PubMed  PubMed Central  Google Scholar 

69. Sabry, M. O., Li, D. & Chan, L. C. Z. A perspective on the selection and design of antimicrobial peptides for bacteria detection. Sens. Bio-Sensing Res. 48, 100794 (2025).

    Google Scholar 

70. Lei, J. et al. The antimicrobial peptides and their potential clinical applications. Am. J. Transl. Res. 11 (7), 3919–3931 (2019).

    PubMed  PubMed Central  Google Scholar 

71. Lin, L. et al. Membrane-disruptive peptides/peptidomimetics-based therapeutics: Promising systems to combat bacteria and cancer in the drug-resistant era. Acta Pharm. Sin B. 11 (9), 2609–2644 (2021).

    PubMed  PubMed Central  Google Scholar 

72. Ridyard, K. E. & Overhage, J. The potential of human peptide LL-37 as an antimicrobial and Anti-Biofilm agent. Antibiot. (Basel) 10(6), 650 (2021).

73. Guha, S. et al. Applications and evolution of melittin, the quintessential membrane active peptide. Biochem. Pharmacol. 193, 114769 (2021).

    PubMed  PubMed Central  Google Scholar 

74. Greber, K. E. & Dawgul, G. Antimicrobial peptides under clinical trials. Curr. Top. Med. Chem. 17 (5), 620–628 (2017).

    PubMed  Google Scholar 

75. Rios, A. C. et al. Alternatives to overcoming bacterial resistances: State-of-the-art. Microbiol. Res. 191, 51–80 (2016).

    PubMed  Google Scholar 

76. Patrulea, V., Borchard, G. & Jordan, O. An update on antimicrobial peptides (AMPs) and their delivery strategies for wound infections. Pharmaceutics 12(9), 840 (2020).

77. Răileanu, M. et al. No country for old antibiotics! Antimicrobial peptides (AMPs) as next-generation treatment for skin and soft tissue infection. Int. J. Pharm. 642, 123169 (2023).

    PubMed  Google Scholar 

78. Duplantier, A. J. & van Hoek, M. L. The human Cathelicidin antimicrobial peptide LL-37 as a potential treatment for polymicrobial infected wounds. Front. Immunol. 4, 143 (2013).

    PubMed  PubMed Central  Google Scholar 

79. Miranda, E. et al. Efficacy of LL-37 cream in enhancing healing of diabetic foot ulcer: A randomized double-blind controlled trial. Arch. Dermatol. Res. 315 (9), 2623–2633 (2023).

    PubMed  PubMed Central  Google Scholar 

80. Svensson, D. & Nilsson, B-O. Human antimicrobial/host defense peptide LL-37 May prevent the spread of a local infection through multiple mechanisms: An update. Inflamm. Res. 74 (1), 36 (2025).

    PubMed  PubMed Central  Google Scholar 

81. Wu, J. et al. A frog Cathelicidin peptide effectively promotes cutaneous wound healing in mice. Biochem. J. 475 (17), 2785–2799 (2018).

    PubMed  Google Scholar 

82. Liu, S. et al. Assessment of antimicrobial and wound healing effects of Brevinin-2Ta against the bacterium Klebsiella pneumoniae in dermally-wounded rats. Oncotarget 8 (67), 111369 (2017).

    PubMed  PubMed Central  Google Scholar 

83. Huang, H. et al. Antimicrobial peptides loaded collagen nanosheets with enhanced antibacterial activity, corneal wound healing and M1 macrophage polarization in bacterial keratitis. Compos. Part. B: Eng. 275, 111283 (2024).

    Google Scholar 

84. Xiao, B. et al. Unlocking the potential of antimicrobial peptides: Cutting-Edge advances and therapeutic potential in combating bacterial keratitis. Bioconjug. Chem. 36 (3), 311–331 (2025).

    PubMed  Google Scholar 

85. SilvaON et al. Repurposing a peptide toxin from Wasp venom into antiinfectives with dual antimicrobial and Immunomodulatory properties. Proc. Natl. Acad. Sci. 117 (43), 26936–26945 (2020).

    ADS  PubMed  PubMed Central  Google Scholar 

86. Li, C. et al. The direct binding of bioactive peptide Andersonin-W1 to TLR4 expedites the healing of diabetic skin wounds. Cell. Mol. Biol. Lett. 29 (1), 24 (2024).

    PubMed  PubMed Central  Google Scholar 

87. Yue, H. et al. Antimicrobial peptide derived from insulin‐like growth factor‐binding protein 5 improves diabetic wound healing. Wound Repair. Regener. 30 (2), 232–244 (2022).

    Google Scholar 

88. Nguyen, H. L. T. et al. The antimicrobial peptide AMP-IBP5 suppresses dermatitis-like lesions in a mouse model of atopic dermatitis through the low-density lipoprotein receptor-related protein-1 receptor. Int. J. Mol. Sci. 24 (6), 5200 (2023).

    PubMed  PubMed Central  Google Scholar 

89. Petkovic, M., Mouritzen, M. V., Mojsoska, B. & Jenssen, H. Immunomodulatory properties of host defence peptides in skin wound healing. Biomolecules. 11(7), 952 (2021).

90. Schroeder, B. O. et al. Paneth cell α-defensin 6 (HD-6) is an antimicrobial peptide. Mucosal Immunol. 8 (3), 661–671 (2015).

    PubMed  Google Scholar 

91. Adyns, L., Proost, P. & Struyf, S. Role of defensins in tumor biology. Int. J. Mol. Sci. 24 (6), 5268 (2023).

    PubMed  PubMed Central  Google Scholar 

Download references

Authors and Affiliations

1. Department of Biomedical Engineering, Meybod University, Meybod, Iran

   Hadi Zare-Zardini

2. Department of Mycology, Pasteur Institute of Iran, Tehran, Iran

   Sima Sadat Seyedjavadi

Contributions

HZZ: Conceptualization, Methodology (Bioinformatic Analysis), Investigation (Cytocompatibility, Wound Healing Assay, Antimicrobial Testing), Data Curation (Computational Dataset), Formal Analysis (Bioinformatic), Supervision, Project Administration, Writing Original Draft Preparation, Writing, Review, and EditingSSS: Methodology (Computational- Machine Learning, Genetic Algorithm), Data Curation (Experimental Data), Formal Analysis (Experimental), Writing, Review , and Editing.

Corresponding author

Correspondence to Hadi Zare-Zardini.

Competing interests

The authors declare no competing interests.

Ethical approval

Not applicable.

Consent for publication

Not applicable.

Clinical trial number

Not applicable.

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions