Genetic imputation of multi-omics profiles for patient stratification and phenotype prediction

• Applicant: Professor Dr. Michael J. Ziller
• Subject Area: Medical Informatics and Medical Bioinformatics ; General Genetics and Functional Genome Biology ; Bioinformatics and Theoretical Biology ; Human Genetics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 553384287

Complex diseases such as cardiovascular or psychiatric illness are highly heritable, but comprise a wide spectrum of symptoms and are clinically heterogenous. The latter is reflected on a genetic level, where GWA studies have unearthed thousands of common genetic associations with these diseases, each associated with extremely small effect size. Consequentially, each individual harbors a private configuration of genetic risk factors for a particular complex disease, resulting in a high level of genetic heterogeneity. It has long been known that clinically defined distinct subgroups of patients within one disease entity also differ in part with respect to their genetic makeup as assessed by e.g. polygenic risk scores. However, these subgroups are frequently not characterized by any common biological basis or shared molecular pathomechanism. We have recently shown that effects of disease-associated genetic variants converge onto distinct cell type specific molecular pathways within distinct genetically stratified subgroups of patients. In particular, we operationalized genotype based imputed transcriptome profiles to show that biological process specific genetic liabilities are not equally distributed across patients. Instead, they define genetically distinct groups of patients, characterized by different expression of pathways, endo- and clinical phenotypes. Here, we propose to build on these findings from imputed transcriptomic profiles and expand this methodology to multi-ome imputation for patient stratification. We hypothesize that the integration of genetically determined variation across multiple molecular layers can be operationalized to define clinically relevant patient biotypes for heritable complex disorders, improving current single-omics models. To test this hypothesis, we will build on the imputation concept and identify distinct patient biotypes based on the individual distribution of genetic factors modulating gene expression, protein abundance and metabolic state. Therefore, we will (i) establish a coherent pipeline to perform tissue specific multi-ome imputation from genotype only data; (ii) implement a novel disease phenotype prediction model from integrated imputed multi-ome profiles using a variational autoencoder neural network and (iii) implement a multi-ome based and endophenotype guided patient stratification strategy utilizing multi-task learning. Finally, we will evaluate and benchmark these new approaches on large, deeply phenotyped cohorts and compare them to traditional single-omics methods. Jointly, this novel method will improve current patient stratification with respect to variance in explained molecular traits, differences in patient level endo- and clinical phenotypes. Thus, this universally applicable approach has the potential to constitute an important component of future personalized medicine concepts.

• DFG Programme: Research Grants