Can artificial neural networks supplant the polygene risk score for risk prediction of complex disorders given very large sample sizes?

TL;DR

This study compares artificial neural networks and polygene risk scores for predicting schizophrenia risk from GWAS data, finding ANNs more sensitive to sample size and potentially more effective with larger datasets.

Contribution

It demonstrates that ANNs outperform PRS in classifying complex disorders like schizophrenia, especially as sample sizes increase.

Findings

ANNs are more sensitive to sample size than PRS.

Larger datasets improve ANN classification performance.

ANNs show promise as an alternative for risk prediction in complex disorders.

Abstract

Genome-wide association studies (GWAS) provide a means of examining the common genetic variation underlying a range of traits and disorders. In addition, it is hoped that GWAS may provide a means of differentiating affected from unaffected individuals. This has potential applications in the area of risk prediction. Current attempts to address this problem focus on using the polygene risk score (PRS) to predict case-control status on the basis of GWAS data. However this approach has so far had limited success for complex traits such as schizophrenia (SZ). This is essentially a classification problem. Artificial neural networks (ANNs) have been shown in recent years to be highly effective in such applications. Here we apply an ANN to the problem of distinguishing SZ patients from unaffected controls. We compare the effectiveness of the ANN with the PRS in classifying individuals by case-control status based only on genetic data from a GWAS. We use the schizophrenia dataset from the Psychiatric Genomics Consortium (PGC) for this study. Our analysis indicates that the ANN is more sensitive to sample size than the PRS. As larger and larger sample sizes become available, we suggest that ANNs are a promising alternative to the PRS for classification and risk prediction for complex genetic disorders.