Second, future polygenic risk prediction models, in addition to common single-nucleotide variants studied in this article, can incorporate rare and low frequency variants43,44 and other complex structural polymorphisms (for example, copy number variations that have been shown to be important for psychiatric disorders45). Given the rapidly evolving technological developments in whole genome and exome sequencing, this is an avenue that will become possible in the very near future. Once identified, these rare variants are likely to have larger effect-sizes (negative selection), and have potential to substantially improve prediction accuracy. Integrating other modalities, including neuroimaging biomarkers and other omics panels such as epigenomics, transcriptomics, metabolomics, and proteomics, is also promising.
