Deep-AFPpred: identifying novel antifungal peptides using pretrained embeddings from seq2vec with 1DCNN-BiLSTM

Area of Science:

• Bioinformatics
• Computational Biology
• Mycology

Background:

• Fungal infections (mycoses) are increasing, exacerbated by COVID-19 and rising antifungal resistance.
• Limited therapeutic options necessitate the search for novel antifungal molecules.
• Antifungal peptides (AFPs) show promise, but natural source identification is costly and slow.

Purpose of the Study:

• To develop a robust in silico model for identifying novel antifungal peptides (AFPs) in protein sequences.
• To create an accessible online prediction server for AFP discovery.

Main Methods:

• Developed Deep-AFPpred, a deep learning classifier using transfer learning with a 1DCNN-BiLSTM algorithm.
• Trained and validated the model on protein sequences to predict AFPs.
• Created a publicly available online server for AFP prediction.

Main Results:

• Deep-AFPpred significantly outperforms existing state-of-the-art AFP classifiers.
• Achieved high precision: approximately 96% on validation data and 94% on test data.
• The online server provides predicted peptides, physicochemical properties, and motifs.

Conclusions:

• Deep-AFPpred offers an efficient and accurate method for identifying novel AFPs.
• The developed online tool facilitates the discovery of potential antifungal agents.
• Identified AFPs can be synthesized and experimentally validated, advancing antifungal drug development.