TOXIFY: a deep learning approach to classify animal venom proteins.

Area of Science:

• Bioinformatics
• Computational Biology
• Toxicology

Background:

• Next-Generation Sequencing and shotgun proteomics generate vast amounts of protein sequence data.
• Traditional methods for empirical toxicity verification struggle to keep pace with data generation.
• Automated identification of toxigenic proteins is essential for efficient toxin discovery.

Purpose of the Study:

• To develop an automated method for identifying venom proteins from sequence data.
• To create a user-friendly software package for toxin identification.
• To improve the speed, efficiency, and accuracy of venom protein classification.

Main Methods:

• Recurrent Neural Networks with Gated Recurrent Units were trained on publicly available datasets.
• Machine learning models were developed for predicting venom protein probability.
• A software package named TOXIFY was created to implement the trained models.

Main Results:

• TOXIFY demonstrates over 20X speed improvement compared to previous methods.
• TOXIFY utilizes significantly less memory (over an order of magnitude less).
• The software achieves higher accuracy, precision, and sensitivity in classifying venom proteins.

Conclusions:

• TOXIFY provides a highly efficient and accurate computational tool for venom protein identification.
• The software facilitates the automation of toxin discovery from large-scale sequencing data.
• TOXIFY represents a significant advancement in bioinformatics for toxicological research.