Compare MAGMA.Celltyping v1.0 vs. v2.0

[bschilder]

@Al-Murphy @NathanSkene

Ran all main analyses in full workflow vignette and test whether:

• Both versions return the same top N significant cell-types for a given GWAS
• P-values are correlated across versions (for the cell-types that overlap since some functions only return significant results).

Now implemented as an optional unit test in test-MAGMA_Celltyping_1.0_vs_2.0.R.

Results summary

• calculate_celltype_associations(EnrichmentMode="Linear")

  • Produces the exact same top 3 enriched celltypes across versions.
  • P-values are 100% correlated.

• calculate_celltype_associations(EnrichmentMode="Top 10%")

  • Produces partially overlapping enriched celltypes across versions, likely due to a bug in the original prepare_quantile_groups code that resulted in uneven numbers of marker genes per celltype.
  • P-values are 75% correlated.

• merge_magma_results

  • Produces partially overlapping top 3 enriched celltypes across versions.
  • P-values are 72-100% correlated.

• calculate_conditional_celltype_associations(EnrichmentMode="Linear", controlTopNcells = 2)

  • Produces partially overlapping top 3 enriched celltypes across versions.
  • P-values are 85-100% correlated.

• calculate_conditional_celltype_associations(EnrichmentMode="Linear", controlledCTs=c("pyramidal CA1","pyramidal SS","interneurons"))

  • Produces partially overlapping top 3 enriched celltypes across versions.
  • P-values are 93-100% correlated.

• calculate_conditional_celltype_associations(EnrichmentMode="Top 10%", controlledCTs=c("pyramidal CA1","pyramidal SS","interneurons"))

  • Produces partially overlapping top 3 enriched celltypes across versions.
  • P-values are 72% correlated.

• calculate_conditional_geneset_enrichment

  • Does not work at all in MAGMA.Celltyping v1.0.0, and therefore cannot be compared.