Poster #24: Deanne Taylor PhD

Deterministic gene-set construction and measurement-matched null design for testing structural organization of curated disease genes in developmental spatial transcriptomics

Deanne Taylor PhD - Children's Hospital of Philadelphia and the University of Pennsylvania Perelman School of Medicine. Ben Stear - Children's Hospital of Philadelphia, Yuanchao Zhang PhD - Children's Hospital of Philadelphia, Taha Mohseni Ahooyi PhD - Children's Hospital of Philadelphia, Aditya Lahiri PhD - Children's Hospital of Philadelphia, Alan Simmons - University of Pittsburgh Medical Center, Jonathan C. Silverstein - University of Pittsburgh Medical Center.

Curated disease gene lists are widely used in enrichment analyses but rarely studied themselves, and set-level tests against spatial transcriptomic data face two confounders: gene sets differ systematically in expression abundance and detection rate, and curated sources overlap in ways that make membership counts uninterpretable. We built a deterministic union of 14 congenital heart disease (CHD) gene sources spanning review tables, a curated database, cohort variant-burden sets, gene-disease validity assertions, and dosage-sensitivity curation. Every symbol was resolved against a dated, checksummed HGNC snapshot, never by approximate matching, yielding 652 current-HGNC genes. We evaluated this registry against developmental human-heart Visium data at 4 and 5 post-conception weeks from two donors, scoring each gene for spatial autocorrelation (Moran's I), regional specificity, and anatomical niche enrichment. Every score was tested against fixed-coordinate permutation nulls, against genome-wide backgrounds matched on abundance and detection rate, and, between gene sets, against an independent matched-null difference contrast, with 100,000 permutations per set-level test across thirteen sources. Two gene subsets sharing no members, those curated as causing CHD in isolation and those causing CHD within a syndrome, differ markedly in their spatial organization: genes associated with isolated CHD are more spatially structured and more regionally restricted early in cardiogenesis. The difference holds within every detection and expression quartile and under partial correlation controlling both. Spatial organization is not specific to curated CHD genes. The 200 genes a published curation panel evaluated but did not retain are as spatially organized as the accepted sets, so the test separates disease classes without distinguishing accepted CHD genes from rejected candidates in the same review. Because panels reject genes for insufficient evidence rather than evidence of absence, this may reflect the comparator's limits as much as the test's.