Integrating cell morphology with gene expression and chemical structure to aid mitochondrial toxicity detection

Abstract

Mitochondrial toxicity is an important safety endpoint in drug discovery. Models based solely on chemical structure for predicting mitochondrial toxicity are currently limited in accuracy and applicability domain to the chemical space of the training compounds. In this work, we aimed to utilize both -omics and chemical data to push beyond the state-of-the-art. We combined Cell Painting and Gene Expression data with chemical structural information from Morgan fingerprints for 382 chemical perturbants tested in the Tox21 mitochondrial membrane depolarization assay. We observed that mitochondrial toxicants differ from non-toxic compounds in morphological space and identified compound clusters having similar mechanisms of mitochondrial toxicity, thereby indicating that morphological space provides biological insights related to mechanisms of action of this endpoint. We further showed that models combining Cell Painting, Gene Expression features and Morgan fingerprints improved model performance on an external test set of 244 compounds by 60% (in terms of F1 score) and improved extrapolation to new chemical space. The performance of our combined models was comparable with dedicated in vitro assays for mitochondrial toxicity. Our results suggest that combining chemical descriptors with biological readouts enhances the detection of mitochondrial toxicants, with practical implications in drug discovery.

More Is Different: Drug Property Analysis on Cellular High-Content Images Using Deep Learning

“…Additionally, MID performs better than CellProfiler and DeepProfiler in task 2, with results consistent with those reported in the literature [21] (Figure 2).…”

“…Multiple mechanisms contribute to mitochondrial toxicity [19] , resulting in various changes in cell phenotype. The alterations in cell morphology, texture, and intensity caused by compounds are strongly correlated with mitochondrial toxicity, suggesting that cell phenotype analysis is a reliable method for predicting mitochondrial toxicity [21] .…”

Using Transcriptomics and Cell Morphology Data in Drug Discovery: The Long Road to Practice

“…Note that not all bioactivity classes might be as well-populated or as easily annotated with a “mode of action”, and hence also here some kind of self-selection has been performed in this study. The general finding, of improved classification performance, is nonetheless consistent with a study cited above, which in particular for novel scaffolds underlined the value of including Cell Painting readouts in predicting biological endpoints.…”

“…46 The predictive value of a readout, given a particular assay setup and data processing pipeline, needs to be validated for every endpoint individually. This has been done for mitochondrial toxicity, 65 using Cell Painting, gene expression, and compound structural information, for 382 chemical perturbants tested in the Tox21 mitochondrial membrane depolarization assay. Mitochondrial toxicants were found to differ from nontoxic compounds in morphological space, and when included in predictive models this combination of features improved model performance on an external test set of 244 significantly, thereby improving extrapolation to new chemical space.…”

Merging Bioactivity Predictions from Cell Morphology and Chemical Fingerprint Models Using Similarity to Training Data

“…This showed that fusing models built on two different feature spaces that provide complementary information were able to improve the prediction of bioactivity endpoints. Previous work has also shown that combinations of descriptors can significantly improve prediction for MOA classification 25,26,15 (using gene expression and cell morphology data), cytotoxicity 16 , mitochondria toxicity 18 and anonymised assay activity 27 (using chemical, gene expression, cell morphology and predicted bioactivity data), prediction of sigma 1 (σ1) receptor antagonist 28 (using cell morphology data and thermal proteome profiling), and even organism-level toxicity 29 (using chemical, protein target and cytotoxicity qHTS data). Thus, the combination of models built from complementary feature spaces can expand a model’s applicability domain by allowing predictions in new structural space.…”