Explainable multi-task learning improves the parallel estimation of polygenic risk scores for many diseases through shared genetic basis

摘要

Author summaryTo prevent or delay the onset of complex diseases, an individual can benefit from knowing which diseases he/she is predisposed to through heredity and which diseases he/she is less susceptible to genetically. The overall genetic risk of a person to a complex disease is quantified using a polygenic risk score (PRS). Traditionally, PRS were developed independently for different diseases using statistical approaches. In this study, we used a multi-task learning approach and trained a deep learning model to simultaneously learn the PRS of many diseases all together. We showed that the new multi-task learning model can provide more accurate estimation of PRS for these diseases than their corresponding single-task learning models that were trained for individual diseases separately. The performance boost by multi-task learning suggests that many complex diseases may share a large number of common genetic risk variants among them, which can contribute to the positive transfer of knowledge during multi-task learning. Many complex diseases share common genetic determinants and are comorbid in a population. We hypothesized that the co-occurrences of diseases and their overlapping genetic etiology can be exploited to simultaneously improve multiple diseases' polygenic risk scores (PRS). This hypothesis was tested using a multi-task learning (MTL) approach based on an explainable neural network architecture. We found that parallel estimations of the PRS for 17 prevalent cancers in a pan-cancer MTL model were generally more accurate than independent estimations for individual cancers in comparable single-task learning (STL) models. Such performance improvement conferred by positive transfer learning was also observed consistently for 60 prevalent non-cancer diseases in a pan-disease MTL model. Interpretation of the MTL models revealed significant genetic correlations between the important sets of single nucleotide polymorphisms used by the neural network for PRS estimation. This suggested a well-connected network of diseases with shared genetic basis.