Application of TensorFlow to recognition of visualized results of fragment molecular orbital (FMO) calculations

TL;DR

This study demonstrates that TensorFlow can effectively recognize characteristic structural patterns in IFIE-maps derived from FMO calculations, aiding in protein structure analysis.

Contribution

The paper introduces a novel application of TensorFlow for recognizing protein structural patterns in IFIE-maps from FMO calculations.

Findings

TensorFlow successfully identified alpha-helix and beta-sheet patterns.

High accuracy in recognizing structural features in new IFIE-map data.

Proven potential for automated protein structure analysis.

Abstract

We have applied Google's TensorFlow deep learning toolkit to recognize the visualized results of the fragment molecular orbital (FMO) calculations. Typical protein structures of alpha-helix and beta-sheet provide some characteristic patterns in the two-dimensional map of inter-fragment interaction energy termed as IFIE-map (Kurisaki et al., Biophys. Chem. 130 (2007) 1). A thousand of IFIE-map images with labels depending on the existences of alpha-helix and beta-sheet were prepared by employing 18 proteins and 3 non-protein systems and were subjected to training by TensorFlow. Finally, TensorFlow was fed with new data to test its ability to recognize the structural patterns. We found that the characteristic structures in test IFIE-map images were judged successfully. Thus the ability of pattern recognition of IFIE-map by TensorFlow was proven.