Steps 3 and 4: gene-level analysis

The Polars concatenation command first combines the per-gene Step 2 files, splits pval_ge into named dynamic-covariate columns, and applies the MAF filter. Step 3 then combines variant-level p-values with ACAT to obtain gene-level p-values. Step 4 calculates q-values separately within each p-value context and calls eGenes at the requested FDR.

The tested commands are available together in the Step 3 + 4 tabs on the CASTIE home page for both container and Pixi installations.

For a real multi-gene dynamic analysis, use step3_gene_pvalue.R to create step3_longformat.txt, then run:

concat_step2_results.py \
  --input-dir=/path/to/step2 \
  --output=/path/to/step3_input.txt \
  --contexts=age,sex,pf1,pf2 \
  --file-pattern='*.txt' \
  --gene-regex='^(?P<gene>.+)_count_cis_window_1000000_0[.]2[.]5[.]7[.]txt$' \
  --maf-min=0.05 \
  --maf-max=0.95

step3_gene_pvalue.R \
  --input=/path/to/step3_input.txt \
  --outdir=/path/to/step3

step4_get_egenes.R \
  --input=/path/to/step3/step3_longformat.txt \
  --outdir=/path/to/step4 \
  --fdr=0.05

Both tab-delimited text and Parquet Step 2 outputs are supported and detected automatically. Select them with --file-pattern='*.txt' or --file-pattern='*.parquet', respectively, and use the same extension in --gene-regex. You can override detection with --input-format=txt or --input-format=parquet.

Step 4 expects Gene, pval_column, and ACAT_p. It writes per-context eGene tables and gene lists, plus context-union, context-only, and shared-context summaries.