Accurate and timely prediction of disease-related traits and identification of the strongest predictors of disease risks are critical for patients, health service providers, and healthcare industries. Early risk prediction can enable prevention and early therapeutic interventions, thus reducing costs of care and improving prognosis, whereas identification of robust predictors of risk or response can lead to novel targets, reduce costs of clinical trials, and improve efficacy of drugs tailored to individual patients. Huge investments are being made to address these goals.

In this project, we aim to improve predictions of complex outcomes using combinations of -omics and clinical/environmental data available from free-public repositories and from proprietary studies conducted by our research partners. This project will focus on the following development objectives:

Establishing familiarity with common and cutting-edge methods developed for high-dimensional predictions.

Developing and applying predictive methods using -omics data in order to outperform traditional clinical models. The specific focus will be on trying to find structure in the data to allow for automatic stratification of patients (or subtyping of diseases) in order to improve predictions.

Incorporating existing medical and/or biological knowledge into predictive models to improve quality of predictions.

Developing an approach using information about multiple related traits to improve prediction of the target trait. To achieve this, we will develop an approach for automatic identification of traits related to the target disease using findings summarized in online databases.

Extending the developed methods for dealing with imbalanced datasets and/or rare features. (Note that population-level studies tend to have a disproportionally low number of “cases”, which will need to be taken into account).