Gene set proximity analysis: expanding gene set enrichment analysis through learned geometric embeddings, with drug-repurposing applications in COVID-19

Abstract

Motivation Gene set analysis methods rely on knowledge-based representations of genetic interactions in the form of both gene set collections and protein-protein interaction (PPI) networks. However, explicit representations of genetic interactions often fail to capture complex interdependencies among genes, limiting the analytic power of such methods. Results We propose an extension of gene set enrichment analysis to a latent embedding space reflecting PPI network topology, called gene set proximity analysis (GSPA). Compared with existing methods, GSPA provides improved ability to identify disease-associated pathways in disease-matched gene expression datasets, while improving reproducibility of enrichment statistics for similar gene sets. GSPA is statistically straightforward, reducing to a version of traditional gene set enrichment analysis through a single user-defined parameter. We apply our method to identify novel drug associations with SARS-CoV-2 viral entry. Finally, we validate our drug association predictions through retrospective clinical analysis of claims data from 8 million patients, supporting a role for gabapentin as a risk factor and metformin as a protective factor for severe COVID-19. Availability GSPA is available for download as a command-line Python package at https://github.com/henrycousins/gspa. Supplementary information Supplementary data are available at Bioinformatics online.

ANDES: a novel best-match approach for enhancing gene set analysis in embedding spaces

“…In 22 out of 31 cases (71% datasets), ANDES set enrichment outperforms the hypergeometric test for KEGG pathway identification (p<7e-4, Wilcoxon rank sum test, Figure 3A). For the rank-based version, we compare ANDES with GSEA [27] as well as a new embedding-based gene set enrichment method, GSPA [29]. Aggregating performance across all 42 datasets in the benchmark, ANDES’ rank-based gene set enrichment significantly improves over both GSEA (p=0.041; Wilcoxon rank sum test) and GSPA (p=0.028; Wilcoxon rank sum test) (Figure 3B).…”

“… (A) Performance comparison between ANDES and hypergeometric test in retrieving annotated KEGG terms using genes that have FDR ≤ 0.05 in each dataset (where there are at least 10 genes that are significantly differentially expressed). (B) Performance comparison between ANDES, GSEA [27], and GSPA [29] in retrieving annotated KEGG terms using the full list of genes (no FDR cutoff), ranked by log 2 (fold change). In both cases, ANDES statistically outperforms other methods, demonstrating the advantage of incorporating gene embedding information using best-match into the gene set enrichment setting.…”

“…Other recent methods that attempt to lessen the dependency of gene set enrichment on existing annotations include Network-Based Gene Set Enrichment Analysis (NGSEA), which reranks the input gene list by incorporating the mean of its network neighbors [37], and Gene Set Proximity Analysis (GSPA), which allows users to supplement gene set annotations using a radius in an embedding space [35]. Unfortunately, we are unable to systematically evaluate NGSEA because it is only available as a web portal, and we cannot change the underlying gene sets used.…”

“…Using the corresponding annotated KEGG pathways as a gold standard, we can then calculate AUPRC for each of the 42 datasets. We compare the results of ANDES against three existing gene set analysis methods: hypergeometric test [26], GSEA [27], and GSPA [29]. We are unable to use NGSEA [28] and EnrichNet [50] for this benchmarking analysis because only web portals are available, with no ability to change the underlying gene sets for a fair comparison across methods.…”

NetAn: A Python Toolbox Leveraging Network Topology for Comprehensive Gene Annotation Enrichments

“…While such leverages of network topology in annotation enrichment analyses have been used before, [5]–[7], few approaches have used networks to disentangle convoluted cellular mechanisms in annotation analyses. If the studied gene set involves genes from two or more moderately independent cellular processes, the signal is obscured, and false negative identifications will be more likely.…”