Predicting individual disease risk for individuals harboring monogenic risk alleles with deep learning

/Abstract Translating genetic information into clinical and public health practice requires a detailed understanding of how genetics influence disease risk. While many monogenic disease risk alleles have been identified, the true level of risk these alleles confer remains unknown for many alleles. Many people with disease risk alleles never develop disease, especially for rare diseases. This disconnect between genetic risk and actual pathogenesis limits the clinical usefulness of pure genetic information. To implement precision medicine requires new methods and techniques that integrate non-genetic factors with genetic risk to produce a complete picture of individual risk. In this proposal, we propose to develop new methods to predict disease risk among individuals who harbor rare monogenic risk alleles, refining the accuracy of genetic diagnostics and increasing the clinical usefulness of genetic information. These methods are based on existing approaches that use deep learning (DL) to impute missing phenotype data. DL methods can leverage information about the context in which diagnoses are seen to learn information about the relationships between diagnoses. This allows them to make accurate predictions about even very rare diseases by analyzing the contexts in which specific diagnoses are seen. At the same time, they can transform these raw data into biologically meaningful signatures, which can be extracted and analyzed to gain insight about disease biology. We will develop and apply these methods in electronic health record (EHR) data, metabolomics data, and proteomics data, using data from the UK Biobank, the NIH All of Us program, the Mount Sinai BioMe Biobank, and the Mount Sinai Million Health Discoveries Program. The result will be robust, validated, and clinically useful prediction methods for disease risk based on DL phenotype imputation, as well as new insights into the biology and genetic architecture of rare genetic diseases. Project Number: 1R01HG013511-01A1 | Fiscal Year: 2025 | NIH Institute/Center: National Human Genome Research Institute (NHGRI) | Principal Investigator: Daniel Jordan | Institution: ICAHN SCHOOL OF MEDICINE AT MOUNT SINAI, NEW YORK, NY | Award Amount: $579,403 | Activity Code: R01 | Study Section: Analytics and Statistics for Population Research Panel A Study Section[ASPA] View on NIH RePORTER: https://reporter.nih.gov/project-details/11045188