Introducing the CSP Analyzer: A novel Machine Learning-based application for automated analysis of two-dimensional NMR spectra in NMR fragment-based screening

R Fino,R Byrne,C.A Softley,M Sattler,G Schneider,G.M Popowicz

doi:10.1016/j.csbj.2020.02.015

Abstract

NMR-based screening, especially fragment-based drug discovery is a valuable approach in early-stage drug discovery. Monitoring fragment-binding in protein-detected 2D NMR experiments requires analysis of hundreds of spectra to detect chemical shift perturbations (CSPs) in the presence of ligands screened. Computational tools are available that simplify the tracking of CSPs in 2D NMR spectra. However, to the best of our knowledge, an efficient automated tool for the assessment and binning of multiple spectra for ligand binding has not yet been described. We present a novel and fast approach for analysis of multiple 2D HSQC spectra based on machine-learning-driven statistical discrimination. The CSP Analyzer features a C# frontend interfaced to a Python ML classifier. The software allows rapid evaluation of 2D screening data from large number of spectra, reducing user-introduced bias in the evaluation. The CSP Analyzer software package is available on GitHub https://github.com/rubbs14/CSP-Analyzer/releases/tag/v1.0 under the GPL license 3.0 and is free to use for academic and commercial uses.

Highlights

In recent times, fragment-based drug discovery (FBDD) has become progressively more important in early-stage drug design
We present a novel approach for very fast analysis of hundreds of 2D nuclear magnetic resonance (NMR) spectra in Fragment-based screening (FBS) based on advanced machinelearning-driven statistical discrimination
To train the machine-learning (ML) model, we focused on developing an input representation, borrowing heavily from advances in computer vision, to enable multi-protein spectral analysis without retraining for each new target of interest

Summary

Introduction

Fragment-based drug discovery (FBDD) has become progressively more important in early-stage drug design. Fragment-based screening (FBS) offers an efficient, rational, way to find small molecule inhibitors. By testing for binding of small fragments, a large chemical space can be tested with fewer molecules than with other approaches. This is due to the higher probability of a suitable binding pocket or position being present with lower complexity molecules [1], leading to higher efficiency (both screening efficiency and ligand efficiency), and elucidating possible starting points for further drug discovery. X-ray crystallography and nuclear magnetic resonance (NMR) can be used to both identify hits and get structural information about the binding. NMR, working in the solution state, and requiring low concentrations, is well placed to detect weakly-

Objectives

Methods

Results

Conclusion