To enhance the screening process for antifungal peptides, the research team was divided into wet-lab and dry-lab groups. The dry-lab experiments encompass data processing, model training, and model validation. The wet-lab experiments serve as complementary validation to the dry-lab results and are designed to assess the safety of antimicrobial peptides. The wet-lab experiments include minimum inhibitory concentration (MIC) determination, hemolysis assays, and mouse experiments.

The model was trained using a dataset comprising more than 30,000 non-antifungal peptide (non-AFP) entries, over 6,000 antifungal peptide (AFP) entries, and fewer than 4,000 MIC data points. The data was randomly divided into training, testing, and validation sets in an 8:1:1 ratio. Two types of models were trained for the antimicrobial peptide screening:

• Qualitative Model: The DeBERTa model was employed for deep learning training, linked to a fully connected layer for classification.
• Quantitative Model: DeBERTa was also used for MIC prediction, trained using mean squared error (MSE) for continuous output.

Model Validation The model was validated using common classification metrics like accuracy, precision, and recall. Regression metrics like MSE and Pearson correlation coefficient were calculated for MIC prediction. Model tuning involved filtering negative sequences to improve performance, increasing prediction accuracy to 99.4%.

The wet-lab experiments served to validate the dry-lab model predictions. The key experiments included:

• Minimum Inhibitory Concentration (MIC) Determination: This experiment was used to assess the efficacy of antifungal peptides in inhibiting fungal growth.

• Hemolysis Assays: These assays tested the safety of antifungal peptides on red blood cells.

• Mouse Model for Skin Infections: A skin infection model in mice was used to assess the therapeutic potential of antifungal peptides.