Data Analysis and Informatics
From instrument to decision: improved decision-making on complex data at scale
Images of drug-perturbed cells harbour a breadth of information about drug effects that can be extracted and interpreted by machine learning methodologies. When combined with genetic perturbations, such assays provide a powerful approach to identify agonists and antagonists of potential targets for therapeutic interventions. Chemical-genetic interactions thereby allow an in-depth characterization of small molecules and their context-dependent effects.
We have established a screening platform using high-throughput pipelines for experimental and analysis workflows which allows us to screen libraries with thousands small molecules. Using a simple staining procedure, termed cellmorph, we extract multiparametric profiles describing overall changes in cellular morphology and cell behavior using automated image analysis to enable a deep phenotypic profiling of small molecules and other perturbations.
Here, we phenotypically measured chemical-genetic interactions between several mutant cell lines carrying single-gene knock outs and several thousand small molecules. Unsupervised clustering and statistical modeling of chemical-genetic interactions revealed promising interactions including synthetic lethality and resistance. We could further show how the phenotypes of mutant cell lines and small molecules can be quantified with machine learning classifiers, allowing a direct scoring and interpretation of drug-induced phenotypes. Importantly, the trained classifiers also efficiently quantified dosage-dependent effects of drugs. Furthermore, we will show how to apply image-based phenotyping to predict drug resistance and sensitivity in patient derived organoids.