Year: 2023
Type: conference-paper
Abstract: Binding between proteins and small molecules plays a crucial role in predicting various biological processes, including metabolic reactions, regulatory mechanisms, and signal transduction pathways. However, binding information is limited, covering only a small fraction of the potential protein-molecule combinations. For example, in the Foodome project, we have data on over 135,000 food-related small molecules, yet only 4.58% of these molecules have known binding annotations with proteins1,2,3. Although deep learning models have been suggested to expedite this identification process, our research demonstrates that state-of-the-art models struggle to generalize to unfamiliar molecular structures, often learning unintended shortcuts. In this presentation, I will introduce AI-Bind4, a novel pipeline that combines network-based sampling strategies with unsupervised pre-training, enhancing binding predictions for proteins and ligands with insufficient or no prior annotations, exemplified by food-related small molecules. AI-Bind predictions were validated via docking simulations and comparison with recent experimental evidence, offering a framework to interpret machine learning prediction of protein-ligand binding by identifying potential active binding sites on the amino acid sequence.
DOI: 10.58647/rexpo.23008
Author: Giulia Menichetti
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4386749722