Logo image
Artificial neural network classification of two-dimensional infrared spectroscopy to facilitate high-throughput screening of allosteric effectors
Dissertation   Open access

Artificial neural network classification of two-dimensional infrared spectroscopy to facilitate high-throughput screening of allosteric effectors

Evan Bradley Schroeder
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2024
DOI: 10.25820/etd.007575
pdf
Thesis - Evan Schroeder (Revisions v2)7.67 MBDownloadView
Open Access Free to read and download

Abstract

Allosteric effectors of enzymes offer many advantages over their orthosteric counterparts, such as increased specificity and selectivity. However, orthosteric pharmaceuticals significantly outnumber allosteric pharmaceuticals when considering FDA approved drugs. One reason for this is the relative ease of screening for potential orthosteric effectors, where all that is required is a binding study. In contrast, potential allosteric effectors must be subject to not only a binding study, but also an assay for altered activity. This significantly increases the time, effort, and resources required to identify potential allosteric effectors. This thesis investigates the use of two-dimensional infrared spectroscopy (2D IR) as a high-throughput screening method for allosteric effectors. Traditionally, 2D IR would be too slow to be considered high-throughput, and so this work attempts to ascertain the minimal amount of data necessary to successfully classify a 2D IR spectrum of a potential effector as either a hit or a miss. To facilitate this classification, I present the use of artificial neural networks. I first demonstrate that neural networks can successfully categorize samples based on their 2D IR spectra, even on reduced datasets. I then show that classification is still possible even when the spectral differences between samples are small and only a single 2D IR datapoint is used. Additionally, I identify trends in terms of which datapoints result in increased classification ability. I finally end with a direct experimental analogue to high-throughput screening of allosteric effectors and show that high accuracy classification is possible on a similar timescale to the state of the art for high-throughput screening of orthosteric effectors.
Optics

Details

Metrics

37 Record Views
Logo image