Equipping Decoy Generation Algorithms for Template-free Protein Structure Prediction with Maps of the Protein Conformation Space

Abstract

A central challenge in template-free protein structure prediction is controlling the quality of computed tertiary structures also known as decoys. Given the size, dimensionality, and inherent characteristics of the protein structure space, this is non-trivial. The current mechanism employed by decoy generation algorithms relies on generating as many decoys as can be afforded. This is impractical and uninformed by any metrics of interest on a decoy dataset. In this paper, we propose to equip a decoy generation algorithm with an evolving map of the protein structure space. The map utilizes low-dimensional representations of protein structure and serves as a memory whose granularity can be controlled. Evaluations on diverse target sequences show that drastic reductions in storage do not sacrifice decoy quality, indicating the promise of the proposed mechanism for decoy generation algorithms in template-free protein structure prediction.

Data Set Augmentation Allows Deep Learning-Based Virtual Screening to Better Generalize to Unseen Target Classes and Highlight Important Binding Interactions

“…This assumption is usually a good one, because the probability that any given ligand will bind to a target’s binding pocket is low, but there are almost certainly ligands labeled as decoys which would, in fact, exhibit binding activity to the targets they are associated with. This is a problem for all machine learning methods in virtual screening, and more intelligent decoy selection is an area of active research …”

“…This is a problem for all machine learning methods in virtual screening, and more intelligent decoy selection is an area of active research. 25 ChEMBL. A set of 50 ChEMBL targets determined to be a suitable benchmark set for 2D fingerprinting methods was compiled by Heikamp and Bajorath.…”

Guiding Protein Conformation Sampling with Conformation Space Maps

“…HEA contains the basic evolutionary ingredients and evolves a fixed-size population of individuals (conformations) for a number of generations. For the memory of the conformation space, we make use of the evolving map we developed previously [26,25] which utilizes low-dimensional representations of protein conformations. The map represents the explored conformation space effectively by storing non-redundant diverse conformations and has considerably small storage requirement compared to a memory which stores all the conformations ever generated.…”