Using Cell Painting and Chemical Data for Small-molecule Bioactivity and Toxicity Prediction

High-content image-based assays have fueled significant discoveries in the life sciences in the past decade, including novel insights into disease etiology, mechanism of action, new therapeutics, and toxicology predictions. The most popular among them is the Cell Painting assay, has been used alone or in combination with other - omics data to decipher the mechanism of action of a compound, its toxicity profile, and many other biological effects. Traditional approaches in drug discovery primarily rely on chemical structure fingerprints, which do not capture biological responses to compounds. By integrating cell morphology data, models offer a more comprehensive view, improving the prediction and accuracy for various bioactivity endpoints and enhancing the interpretability of models. We present a comprehensive approach to improving the prediction of drug toxicity and understanding the underlying mechanisms by integrating chemical and biological data. We first demonstrate the advantages of combining cell morphology data from the Cell Painting assay, gene expression data, and chemical structural information to develop machine learning models that accurately predict mitochondrial toxicity and show comparable performance with dedicated in vitro assays. The applicability domain of machine learning models trained on structural fingerprints for the prediction of biological endpoints is often limited by the lack of diversity of chemical space of the training data. We developed similarity-based merger models which combined the outputs of individual models trained on cell morphology (based on Cell Painting) and chemical structure (based on chemical fingerprints) and the structural and morphological similarities of the compounds in the test dataset to compounds in the training dataset. The similarity-based merger models improved predictions across a wide range of biological assays from ChEMBL, PubChem and the Broad Institute. Cell Painting assays, however, faced a significant hurdle in the industry: complex image-based features needed to be more interpretable. To bridge the gap between high-dimensional morphological features and biologist-interpretable phenotypes, we introduce an algorithm that maps Cell Painting features to various readouts from the Cell Health assay. The BioMorph space revealed the mechanism of action for individual compounds, including dual-acting compounds such as emetine. This approach therefore offers a biologically relevant way to interpret the cell morphological features derived using software such as CellProfiler and to generate hypotheses for experimental validation. In summary, our work demonstrates the potential of combining chemical and biological data for enhanced prediction of drug safety endpoints with practical implications in drug discovery and predictive toxicology.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2024

Author dc:creator

• Seal, Srijit

Advisor dc:contributor.advisor

• Bender, Andreas

• Assay
• Cell Painting
• Cheminformatics
• Machine Learning
• Toxicity

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/efecc771-c5d5-4f0a-8778-b1d0cdcefc37/download
• https://creativecommons.org/licenses/by/4.0/

Language dc:language

eng

Author Identifier

0000-0003-2790-8679

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/375315