Machine learning analysis of psychiatric diseases
Principal Investigator:
Jason Moore
Approved Research ID:
51709
Approval date:
November 6th 2019
Lay summary
An important goal of psychiatric disease research is to develop genetic predictors of susceptibility. The overarching goal of this 3-year project is thinking about and approaching the genetic analysis of psychiatric diseases from a complex systems point of view. Genetic variants include common and rare variants and both are expected to play an important role in common diseases. In contrast to common variants, rare variants are difficult to analyze because there are so few individuals in the population that have the rare allele. This makes it difficult to statistically compare cases and controls. One answer to this has been to collapse multiple rare variants across a genomic region, but the current approaches are simplistic in that they treat all variants as being equal and additive while at the same time ignoring their spatial context. We will develop a biologically-inspired approach to the analysis of rare variant associations methodology that can identify optimal subsets of rare variants and, at the same time, identify the optimal way to collapse them into new common variants that can be used in genetic association studies. We will then analyze combinations of the resulting collapsed variants together with actual common variants for their association with psychiatric diseases. This will be done via non-parametric and genetic model-free machine learning approaches that will enable us to embrace the complexity of the relationship between genomic variation and phenotypic variation.