The computational part of the MORPHology-based Endocrine DisrUptor Screening (Morpheus) project aims to develop deep learning models to predict the effects of substances and identify characteristic fingerprints based on morphological and molecular input data.
The BMBF-funded Morpheus project aims to enable the efficient identification and characterization of endocrine-like substances with harmful effects on humans and the environment. Those substances are often studied in time-consuming experiments involving animals raising ethical issues. Our partner (BfR and FMP) and us (UdS) propose to study them using in vitro methods such as cell painting assay combined with in silico modelling. Cell painting assay data reveals morphological changes in the cell, and thus, how the substances impact cell viability. Such data can nowadays be produced with high throughput (e.g. CIL). This information, after being converted to numbers as morphological fingerprints, is easily interpretable by computers. Thus, we propose to develop machine learning and deep learning methods that leverage information from morphological fingerprints and combine them with structural information, e.g. structural fingerprints. Such tools will help to predict substance effects and to identify characteristic fingerprints of hazardous compounds. These models will be iteratively trained and validated based on freely available data (CIL), as well as applied and optimized on new profiles successively generated by our project partners from BfR and FMP.
Finally, the generated in silico models will be applied to effectively search large substance libraries, which will be experimentally verified.