DeepGWAS: Enhance GWAS Signals for Neuropsychiatric Disorders via Deep Neural Network

Abstract

Genetic dissection of neuropsychiatric disorders can potentially reveal novel therapeutic targets. While genome-wide association studies (GWAS) have tremendously advanced our understanding, we approach a sample size bottleneck (i.e., the number of cases needed to identify >90% of all loci is impractical). Therefore, computationally enhancing GWAS on existing samples may be particularly valuable. Here, we describe DeepGWAS, a deep neural network-based method to enhance GWAS by integrating GWAS results with linkage disequilibrium and brain-related functional annotations. DeepGWAS enhanced schizophrenia (SCZ) loci by ~3X when applied to the largest European GWAS, and 21.3% enhanced loci were validated by the latest multi-ancestry GWAS. Importantly, DeepGWAS models can be transferred to other neuropsychiatric disorders. Transferring SCZ-trained models to Alzheimer’s disease and major depressive disorder, we observed 1.3-17.6X detected loci compared to standard GWAS, among which 27-40% were validated by other GWAS studies. We anticipate DeepGWAS to be a powerful tool in GWAS studies.

Functional characterization of gene regulatory elements and neuropsychiatric disease-associated risk loci in iPSCs and iPSC-derived neurons

“…SCZ GWAS SNPs were initially selected if supported by at least two studies from four SCZ studies 8,46,47,72 . A significance threshold of P < 5e-8 was applied to the three GWAS studies 8,46,47 , while for DeepGWAS 72 , a posterior probability greater than 0.5 was utilized.…”

“…SCZ GWAS SNPs were initially selected if supported by at least two studies from four SCZ studies 8,46,47,72 . A significance threshold of P < 5e-8 was applied to the three GWAS studies 8,46,47 , while for DeepGWAS 72 , a posterior probability greater than 0.5 was utilized. SNPs in linkage disequilibrium with these initial variants (r 2 > 0.6) were also incorporated from the TOP-LD database 73 , focusing on the EUR population, resulting in a comprehensive collection of 25,999 SNPs.…”

S-BEAM: A Semi-Supervised Ensemble Approach to Rank Potential Causal Variants and Their Target Genes in Microglia for Alzheimer’s Disease

“…We first identified candidate causal variants by integrating results from the latest AD GWAS study with functional annotations using our recently developed DeepGWAS method (Li et al, 2019). Specifically, we started from the GWAS meta-analysis of AD performed by Bellenguez et al recently, testing the association between 21,101,114 variants and the disease in 111,326 clinically diagnosed or “proxy” AD cases and 677,633 controls.…”

“…Variants overlapping with a functional element were assigned a value of 1 for that annotation feature and 0 otherwise. Additionally, we obtained CADD-PHRED scores, FATHMM-XF scores, and eQTL data from external databases (Li et al, 2019). Another two features taken into account by DeepGWAS are the LD score (with any variant) and LD score with variants reported by Bellenguez et al to be associated with AD (i.e., p-value < 5e-8).…”

Machine Learning to Advance Human Genome-Wide Association Studies

“…Another method is DeepGWAS which uses a 14-layer deep neural network to enhance power in GWAS signals without increasing the sample size, by assigning unequal a priori probability for each SNP involvement in disease leveraging linkage disequilibrium information and brain-related functional annotations. DeepGWAS was developed particularly for psychiatric diseases, starting with schizophrenia and outperformed XGBoost and logistic regression methods [70]. COMBI [71] and DeepCOMBI [72] also have built-in ML-based variant prioritisation functions which are discussed in more detail below.…”