Predicting the three-dimensional structure of a protein from its amino acid sequence remains a formidable computational challenge. Small proteins have been successfully folded into their native structures using molecular dynamics, yet simulation-based approaches remain impractical for most proteins. Recent methods based on co-evolution inference have shown remarkable success at predicting protein folds, yet those methods can only be used when large numbers of ortholog sequences are available.<br/><br/>Since the native structure of a protein is the result of a physical process primarily determined by the sequence of the protein itself, one can imagine that protein folding can be learned in a way similar to how a complex game such as chess or Go can be learned. With that general idea in mind, we are developing and training deep neural networks to perform two main tasks: (1) generating backbone and side-chain dihedral angles corresponding to plausible native conformations of a protein, and (2) ranking protein conformations according to how similar they are to an unknown native structure. In this talk, I will be presenting networks we have developed for each task (1 & 2) and will be discussing our recent progresses on assembling them into an accurate protein structure prediction tool.<br/>

Modeling and prediction of protein structure using deep neural networks

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.