Prediction of new neurotoxins from the Latrotoxin family in Spiders using Deep Neural Networks and augmented database

• Grant number: 24/07032-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: October 01, 2024
• End date: September 30, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Milton Yutaka Nishiyama Junior
• Grantee: Gustavo Akio Honda
• Host Institution: Instituto Butantan. Secretaria da Saúde (São Paulo - Estado). São Paulo , SP, Brazil

Abstract

The study of the composition and physiological action of arachnids venom toxins is of great value to the medical and biotechnological field, serving as potential active compounds for new drugs development. In recent decades, new proteins have been obtained from non-model organisms or generated by generative computational models, being functionally annotated by traditional homology approaches. However, accurate and higher quality annotation approaches that allows evaluating its correct folding and determining its functional activity remains a challenge. The molecular evolutionary route and composition of these venom toxins present a high complexity, with studies that indicate conservation in relation to their ancestral genes, and an interrelationship of factors such as gene duplication, alternative splicing, and post-translational modifications presenting mechanisms that enable the diversification of the divergent effectiveness of the physiological action in relation to the ancestral gene, generating venom diversity. The neurotoxins families are the subject of this study, specially the Latrotoxins. We aims to establish a pipeline using a model based on deep neural networks to expand the representation of proteins in this family and classify the action of the compounds in order to facilitate the identification of proteins that have not yet been classified. or hipothetical. Aiming to generate new proteins to expand training banks and establish a classification process regarding their neurotoxic functionality, we intend to evaluate and use the latest Deep Neural Network technologies, such as adversarial generative neural network models, as well as statistical methods such as reconstruction of ancestral sequences. Further we will construct a database of functional proteins from the neurotoxin and latrotoxin families to represent the functional adjustment of proteins belonging to the respective family, which will be used in our model, using an algorithm of large protein language models. Being a relatively new and growing area, we intend to establish metrics to evaluate the quality of sequences and their functional activity in order to boost research in protein engineering, serving as a reference for generative models of protein sequences and helping to select active variants for experimental tests, providing discoveries in both the biological and computational areas and improving the associate projects under development by our group.