‘The Role of Machine Learning and Cell-Painting in Drug Development’ was the opening presentation for my master thesis supervised by the TU Darmstadt and the University of Cambridge. The project was embedded in computational biochemistry and data science whilst the Darmstadt group did experimental research. Therefore the aim of this presentation was to introduce the work group in Darmstadt to the necessary research and the methods used. Hence, methods like machine learning, cell painting and what can hopefully be achieved by combining the two is briefly outlined which means that the presentation is suitable for scientists who are not experts on the topic.
Dr. Andreas Bender was my supervisor at the University of Cambridge during the time I gave that presentation.
Prof. Dr. Schmitz was my supervisor at the Technische Universitaet Darmstadt during the time I gave that presentation.
Picture made by Thomas Shafee
Image made by Recursion Pharmaceuticals
Image made by angellodeco | Fotolia
Figure made by Broad Institute
Logo of CellProfiler-Software
Image made by CJ Haughey
Picture made by Emilie Bess
Schema made by Embention
Picture made by May Lee
Original publication, for a large phenotypic profiling assay, released from Broad Institute. This paper explains the experimental methods used to undertake the assay and the purpose and possible applications of this rather large data set. It is a very detailed paper for very interested readers with a large methods section that gives every information needed to replicate the results.
This publication discusses a cell painting assay released from Broad Institute. It is a brief summary of how the image data is first obtained experimentally, a list of the fluorescent dyes is given and afterwards the CellProfiler Software is referenced for being the tool to derive numerical features with. A short overview over potential uses is also given.
A phenotypic screening assay for glucocorticoid receptors is here repurposed to predict protein activity in seemingly unrelated assays. From this assay they built a machine learning model, helped increase their drug target hit rates by up to 250 times in experiment.