Approved Research

Prediction of drug response using pharmacogenomics-driven machine learning approach

Lay summary

Precision medicine tailors diagnoses and treatment options on an individual patient basis. This is a paradigm shift from choosing a treatment based on aggregate therapeutic efficacy (i.e., percentage of patients achieving the desired therapeutic benefit in clinical trials). Individualized treatment is greatly needed because not all patients will respond to a particular drug, and drug response varies among patients. Reduction of costs in assaying patients' genetic makeup (genome) among other biological measures has accelerated advances in precision medicine. However, there are only a few well-characterized systemic diseases for which dramatically improved treatment outcomes have been achieved by using genomics data in routine clinical settings. To expand the benefits of individualized medicine to include other widely prevalent complex diseases (e.g., mental health disorders, migraines etc.), a key question to address is: "Can a clinician's assessment of disease severity, augmented with patient-specific biological measures help identify a therapeutic agent (e.g., drug) and dose with the highest likelihood of achieving the desired therapeutic benefit?" The answer to that question requires "intelligence" from high-quality data, rich clinical insights/annotations from domain experience (e.g., physicians, biologists), and analytical approaches to combine heterogeneous patient-specific measures to generate "artificial intelligence." The aims of this study are to identify and validate genetic variants by data-driven pharmacogenomic approach on drug responses and adverse events, using UK Biobank data. Developing  this  systematic  prediction  framework  that  is  optimized  for pharmacogenetic  assessments  can  help  each  individual  achieve  an  optimal therapeutic response, avoid therapeutic failures, and minimize drug-induced toxicity. The estimated duration of our project is more than 3 years.

Scope extension: We aim to develop a pharmacogenomics-driven machine learning approach that can predict the clinical prognosis for drug treatment. Expanding on this scope, it is important to consider that the prognosis in pharmacogenomics is often influenced by a multitude of phenotypic factors. These can range from age, underlying conditions such as hypertension, diabetes, renal failure, heart failure, to blood tests, lifestyle patterns, and more. Recognizing the significant impact of these factors on clinical prognosis, our research intends to incorporate them comprehensively into the prediction model. We plan to utilize diverse artificial intelligence techniques to accurately forecast the response to drug treatment, taking into account this broad spectrum of influential variables.