Titterton, K. | Search bioRxiv

bioRxiv2026

Cell Painting and machine learning for mechanistic profiling of hepatotoxic botanicals and their constituents

BACKGROUND Botanicals are widely consumed as herbal remedies and plant-derived products, but their complex and variable chemical composition presents major challenges for safety assessment. Although hepatotoxicity has been reported for several botanicals, the constituent compounds responsible and the underlying mechanisms often remain poorly characterised. OBJECTIVES We aimed to determine whether Cell Painting could identify candidate constituent drivers of botanical bioactivity and generate hypotheses about potential mechanisms of action in primary human hepatocytes. METHODS We profiled 13 botanical extracts and 85 known constituents across multiple concentrations using Cell Painting alongside assays of metabolic activity and membrane integrity. To interpret morphological phenotypes, we trained machine-learning models using a Cell Painting reference atlas of >1,000 chemicals linked to ToxCast/Tox21 molecular and cellular bioactivity annotations. RESULTS Cell Painting detected bioactivity in 11/13 botanical extracts and 46/85 constituents, compared with 9/13 extracts and 40/85 constituents in the cell-health assays, and generally detected activity at lower concentrations. For 9/13 extracts, at least one constituent was active at a lower concentration than the corresponding botanical extract, supporting the contribution of a subset of potent constituents to mixture-level responses. Aristolochia and goldenseal were the most potent extracts across Cell Painting and cell-health assays and contained highly active constituents. Reference-atlas models generated specific mechanistic hypotheses, including cytotoxicity-, endocrine-, and genotoxicity-associated responses for several botanicals and constituents. CONCLUSION These results support Cell Painting combined with machine learning as a sensitive approach for characterising cellular responses to botanicals, prioritising candidate constituent drivers of hepatotoxicity-associated phenotypes, and generating testable hypotheses about potential mechanisms of action.

pharmacology and toxicology↗

bioRxiv2025

Small molecule bioactivity benchmarks are often well-predicted by counting cells

Phenotypic profiling methods, such as Cell Painting and gene expression, have been widely used to predict compound bioactivity, often showing improvement over predictive models based on chemical structures alone. We discovered that a large subset of assays in widely-used benchmark datasets either directly relate to cell health and cytotoxicity or are assays intending to capture a more specific phenotype but whose active compounds impact cell count, while inactives do not. As a result, counting cells can achieve similar predictive performance as Cell Painting or gene expression data. Filtering benchmarks to include only assays relating to protein targets reveals that Cell Painting can capture information that cannot be predicted by mere cell counting. We re-evaluated three benchmark datasets used with Cell Painting data and observed that, in many cases, cell count models produced an AUC comparable to models using the full Cell Painting profiles. However, in protein-target-specific benchmarks across 17 distinct protein targets, Cell Painting features demonstrated unique predictive power, outperforming mean balanced accuracy from cell count models with a relative improvement of 19.6%. We propose five practical recommendations for benchmarking machine learning models for predicting bioactivity, including using cell count as a baseline feature. Although multi-class classification applications (such as matching samples based on their morphological profile) are less likely to be predictable by cell count than bioactivity benchmarks, these recommendations are broadly applicable to machine learning for drug discovery.

bioinformatics↗