Mechanistic Task Groupings Enhance Multitask Deep Learning of Strain-Specific Ames Mutagenicity

Abstract

The Ames test is a gold standard mutagenicity assay that utilizes various Salmonella typhimurium strains with and without S9 fraction to provide insights into the mechanisms by which a chemical can mutate DNA. Multitask deep learning is an ideal framework for developing QSAR models with multiple end points, such as the Ames test, as the joint training of multiple predictive tasks may synergistically improve the prediction accuracy of each task. This work investigated how toxicology domain knowledge can be used to handcraft task groupings that better guide the training of multitask neural networks compared to a naïve ungrouped multitask neural network developed on a complete set of tasks. Sixteen S. typhimurium ± S9 strain tasks were used to generate groupings based on mutagenic and metabolic mechanisms that were reflected in correlation data analyses. Both grouped and ungrouped multitask neural networks predicted the 16 strain tasks with a higher balanced accuracy compared with single task controls, with grouped multitask neural networks consistently featuring incremental increases in predictivity over the ungrouped approach. We conclude that the main variable driving these performance improvements is the general multitask effect with mechanistic task groupings acting as an enhancement step to further concentrate synergistic training signals united by a common biological mechanism. This approach enables incorporation of toxicology domain knowledge into multitask QSAR model development allowing for more transparent and accurate Ames mutagenicity prediction.

Introduction to the Special Issue: AI Meets Toxicology

“…The methods described in this special issue cover a wide range of AI methods ranging from expert systems, ,, over similarity measures including read-across methods, − to classical machine learning such as random forests (RF), support vector machines (SVM), and artificial neural networks (ANN) − ,, to deep learning (DL) methods ,,− , including equivariant neural networks, deep generative models, and even large language models . In addition to models relying purely on the chemical structure, there is a notable trend of bringing in additional modalities to improve or inform predictive models. , In the following, we provide an overview of the AI approaches used in the publications contained in the SI.…”

“…Their findings suggest that MIL could supplement existing models, especially for metabolically activated xenobiotic mutagens in Ames tests with S9. Lui et al have also approached Ames mutagenicity, investigating the utility of multitask deep learning and toxicology domain knowledge for task grouping in developing QSAR models. Their findings suggest that grouped multitask neural networks, informed by mechanistic task groupings, outperform both ungrouped neural networks and single-task models, thereby showcasing the benefits of blending mechanistic and data-driven modeling for toxicological model development.…”

“…The tasks and problems spanned by the proposed methods are also highly diverse and range from classical quantitative structure–activity/property relationship (QSAR/QSPR) problems, such as the prediction of drug-induced liver injury (DILI), , Ames mutagenicity, , or acetylcholinesterase (AChE) inhibition, to categorization of chemicals, generating transcriptomic profiles and classifying parts of regulatory documents . In their paper, “Transparency in Modeling through Careful Application of OECD’s QSAR/QSPR Principles via a Curated Water Solubility Data Set”, the authors revisited the five Organisation for Economic Co-operation and Development (OECD) principles for QSAR models.…”

“…Most of the presented methods perform prediction tasks on molecules (for exceptions, see paragraphs below) and, thus, have to use representations of molecules. While for data handling, the simplified molecular-input line-entry system (SMILES) representation is often used, the predominant representations of molecules for modeling are still extended connectivity fingerprint (ECFP) or Morgan fingerprints. − ,− Several publications use graph neural networks that operate on the molecular graph. , Aside from the chemical structure, there is a growing tendency to incorporate biological characterizations and read-outs, for example, via cell morphology , or transcriptomics . The utilization of diverse representations, ranging from molecular structures to biological features, enhances the predictive models showcased in this section and could improve the comprehensive understanding of toxicological properties.…”

Machine learning in toxicological sciences: opportunities for assessing drug toxicity

“…A study using neural network ML models employed splitting of strain tasks. Multitask neural networks were more accurate than single-task neural networks, and grouped multitask neural networks were superior, most likely because the latter was grouped rationally by mutagenic and metabolic mechanisms (Lui et al, 2023). However, complete automation and reliance on the ML models may not produce the most accurate results.…”