Towards precision medicine: Deep learning of drug response patterns based on patient-derived blood cancer samples

• Applicant: Dr. Tim Heinemann
• Subject Area: Epidemiology and Medical Biometry/Statistics ; Bioinformatics and Theoretical Biology
• Term: from 2017 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 389640585

Drug responses measured in patient-derived cancer tissues may predict therapeutic outcomes and enable the physiological assessment of drug action. The quality of those predictions depends on the accuracy by which the cancer cells can be identified, which in turn relies on the identification of disease- and patient-specific biomarkers. Additionally, optimal ex vivo drug response measurements should be sensitive, have a single-cell resolution, and enable the testing of many drugs from small patient samples. In the proposed research project, I will apply state-of-the-art deep neural networks (DNN) for the automated image-based diagnostic of ex vivo patient samples and prediction of the patient’s response to drug treatment. The machine learning (ML) architecture will be trained and validated on – yet unpublished – patient derived primary blood cancer samples, which were screened for their sensitivity to over 140 clinically approved drugs. By perturbing the best-performing DNNs I aim to determine predictive morphological and categorical features, which will help to characterize the patient specific disease state, infer the ex vivo drug response, and predict clinical response to the tested treatments. Overall, the combination of deep learning and image-based drug screening may offer an attractive platform for the fast and personalized identification of clinically effective therapies, for haematological malignancies and beyond.

• DFG Programme: Research Fellowships
• International Connection: Switzerland
• Host: Professor Dr. Berend Snijder