2023-07-12

I presented a high level summary of the methods we are developing.

Neuropsychiatric disorders (NPD) are complex, heterogeneous disorders. Matrix factorization is a method that has been used to identify the shared and distinct factors of NPDs. Truncated SVD and genomic structural equation modeling (SEM) has been used earlier for multi-phenotype analysis. We are exploring different convex methods to obtain a low rank approximation of the input matrix before doing a PCA. Convex methods are reproducible. Currently, we are looking at 4 different methods for nuclear norm regularization to obtain the low rank approximation. We expect the low rank approximation will provide a more accurate estimate of the SNP’s true effect on a phenotype, cell type or other biological trait.

As of now, we hope to use the z-score and standard error for our model. GWAS data does not have to be limited to only patients with a diagnosis of BD, MDD or SZ as summary statistics from any phenotype will be modeled appropriately.

Current main things to consider in regards to the model are:

• What is the normalisation method used in each of the GWAS summary statistics?
• Is there sample overlap between the studies?

Shane will follow-up with:

• Researching the normalisation method used in the selected studies.
• Researching the cohorts used / sample overlap between studies.