Generalizable biomedical informatics strategies for predictive modeling of treatment response

点击翻译按钮获取中文摘要

Identification of patients with poor and favorable treatment response prior to therapy administration is invaluable for improving patient survival and disease management. We propose to build an open-source scalable generalizable method that would assist experimentalists and clinicians on assessing patient's risk of developing therapy resistance and would establish a foundation for our long-term goal to build a platform for patient-centric clinical decision making, personalized therapeutic advice, and disease management. We propose to develop a generalizable versatile bioinformatics paradigm that will use patient molecular profiles to PREDICT their Therapy Response, PREDICTTR, which combines network analysis, statistical modeling, and ensemble machine learning in a unique innovative way that allows accurate elucidation of complex multi-level relationships that govern treatment response. The objective of our proposed approach is two-fold: (i) uncover molecular markers and valuable candidates for therapeutic intervention, which can potentially be targeted to preclude or overcome resistance; and (ii) predict patient's response to therapy administration, which holds a long-term promise to improve disease outcome and reduce the cost of unnecessary and ineffective treatments. Motivated by increasing cases of treatment resistance in oncology, we will apply our algorithm to elucidate (i) response to androgen targeting in prostate cancer and (ii) response to standard-of-care chemotherapy in acute myeloid leukemia. We will disseminate our approach through a web-based decision- making tool, which will be implemented through a Hadoop-oriented solution to (i) broaden its practical impact and (ii) establish clinical utility. Taken together, this multi-task resource is a unique innovative effort of its kind in the therapeutic resistance space with a direct broad impact on personalized therapeutic advice and disease management. Even though we will train our model in prostate cancer and acute myeloid leukemia, our approach can be easily and broadly applicable to other therapies and diseases. This effort will be led by an Early Stage Investigator, Antonina Mitrofanova (PI) who has extensive training and expertise in biomedical informatics and big data analytics. Her collaborative team includes Dr. Shantenu Jha (Rutgers, co-I) who is an expert in distributed systems and will advise on Hadoop development and validation; Dr. Shridar Ganesan (Rutgers, co-I) who will provide clinical and sequencing patient data and incorporate the utilization of our method into the Rutgers CINJ Molecular Tumor Board; Dr. Isaac Kim (Rutgers, co-I) who will provide additional data for validation in prostate cancer; Dr. Christopher Hourigan (NHLBI , NIH, Significant Collaborator), who will provide data for clinical validation in acute myeloid leukemia and is committed to test our web-based portal; and Dr. Scott Parrott (Rutgers, co-I), who is an expert in statistical analysis and will consult on power calculations and multiple testing corrections.