A Deep Learning Approach with Data Augmentation to Predict Novel Spider Neurotoxic Peptides

Area of Science:

• Biochemistry
• Bioinformatics
• Pharmacology

Background:

• Neurotoxic peptides in spider venoms are valuable pharmaceutical resources.
• Identifying these peptides is challenging due to limited data and laborious traditional methods.
• Deep learning offers a promising alternative for peptide identification.

Purpose of the Study:

• To develop and validate a data augmentation method for improving deep learning-based neurotoxic peptide identification.
• To discover novel neurotoxic peptides from *Callobius koreanus* spider venom using the developed model.
• To assess the pharmaceutical potential of newly identified peptides.

Main Methods:

• Implemented a convolutional neural network (CNN) model for neurotoxic peptide recognition.
• Augmented existing neurotoxic peptide data from the UniProt database.
• Trained and compared CNN models using both augmented and unaugmented datasets.
• Utilized Basic Local Alignment Search Tool (BLAST) for sequence homology analysis.
• Validated neuromodulatory effects of selected peptides on SH-SY5Y neuroblastoma cells.

Main Results:

• The CNN model trained with augmented data achieved high performance (accuracy: 0.9953, F1-score: 0.9953).
• Identified 275 putative neurotoxic peptides from *Callobius koreanus*, including 252 novel sequences.
• Four novel peptides demonstrated neuromodulatory effects on human neuroblastoma cells.

Conclusions:

• The proposed data augmentation method significantly improves deep learning model performance for neurotoxic peptide identification.
• This approach facilitates the discovery of novel functional peptides from limited biological data.
• The identified peptides hold potential for pharmaceutical applications in neurodegenerative diseases and pain management.