NucleoFind: A Deep-Learning Network for Interpreting Nucleic Acid Electron Density

TL;DR

NucleoFind is a deep-learning tool that accurately identifies nucleic acid components in electron density maps, speeding up molecular model building.

Contribution

Introduces NucleoFind, a novel deep-learning approach for accurate nucleic acid electron density interpretation.

Findings

NucleoFind identifies 86% of phosphate atoms, 90% of sugar atoms, and 90% of base atoms in electron density maps.

The method enables faster and more accurate nucleic acid model building after molecular replacement.

Abstract

Nucleic acid electron density interpretation after molecular replacement remains a difficult problem for computer programs to deal with. Programs tend to rely on time-consuming and computationally exhaustive searches to recognise characteristic features. We present NucleoFind, a deep- learning-based approach to interpreting and segmenting electron density. Using an electron density map from X-ray crystallography, the positions of the phosphate group, sugar ring and nitrogenous base group can be predicted with high accuracy. On average, 86 % of phosphate atoms, 90 % of sugar atoms and 90 % of base atoms are located by the deep learning model in maps produced following protein molecular replacement. The wealth of context these predicted maps provide can then be used to automatically build more accurate and complete nucleic acid molecular models in a short amount of time.