Artificial intelligence improves drug testing in patient samples for personalised medicine
A recent "Blood Cancer Discovery" paper by the Snijder group (IMSB) in collaboration with the Medical University of Vienna, CeMM, and Exscientia shows that deep learning of cell morphologies improves the clinical treatment recommendations obtained by drug screening in patient samples.
Finding effective treatments for patients with aggressive leukaemias and lymphomas remains an urgent clinical challenge. To address this, the group led by Prof. Dr. Berend Snijder at the Institute of Molecular Systems Biology is pioneering a platform to rapidly measure the response to hundreds of drugs in small patient biopsies. The platform, called Pharmacoscopy, combines automated microscopy and single-cell image analysis to see the drug response of each cell in the biopsy. In a previously published interventional clinical study by the same consortium, Pharmacoscopy-recommended treatments clinically outperformed the previous treatments given to 56 patients suffering from aggressive blood cancers (Kornauth et al., Cancer Discovery, 2022).
Now, the consortium reports that such ‘ex vivo drug screens’ on 390 patient samples are technically, biologically, and clinically improved when using artificial intelligence to analyse the microscopy images. The team developed a deep learning approach that analyzes cell morphologies to classify the one billion imaged patient cells as either healthy or cancerous. Remarkably, when re-analyzing the previous clinical trial data, the authors find that the deep learning approach recommended treatments that led to further improved clinical responses.
The study is published in the journal Blood Cancer Discovery of the American Association for Cancer Research with shared-first authors Dr. Tim Heinemann and Dr. med Christoph Kornauth.
Link to the paper in external pageBlood Cancer Discovery