Using recurrent neural networks to predict aspects of 3-D structure of folded copolymer sequences

TL;DR

This paper develops recurrent neural network methods to predict 3-D structural aspects of folded copolymer sequences, focusing on hydrophobicity and long-range interactions, to improve understanding of protein-like sequences.

Contribution

It introduces a recurrent neural network approach tailored for long sequences and long-distance interactions in protein-like structures, using hydrophobicity as the primary feature.

Findings

Neural networks can predict 3-D structural features of copolymer sequences.

Recurrent neural networks effectively handle long sequences with distant interactions.

Hydrophobicity encoding suffices for initial structural predictions.

Abstract

The neural network techniques are developed for artificial sequences based on approximate models of proteins. We only encode the hydrophobicity of the amino acid side chains without attempting to model the secondary structure. We use our approach to obtain a large set of sequences with known 3-D structures for training the neural network. By employing recurrent neural networks we describe a way to augment a neural network to deal with sequences of realistic length and long-distant interactions between the sequence regions.