scripts

Imported from Yuanwei-source/OmniQuant-RNA (workflow/scripts/AGENTS.md). Install upstream with npx skills add Yuanwei-source/OmniQuant-RNA --skill scripts. Copyright stays with the author.

workflow/scripts/ — Analysis Scripts

OVERVIEW

19 standalone scripts (14 Python + 5 R) invoked by Snakemake rules. Python handles data aggregation, namespace construction, and manifest generation. R handles all statistical analysis (DESeq2 DEA, RRA/CCT consensus, visualization).

LANGUAGE SPLIT

PYTHON CONVENTIONS

Boilerplate

#!/usr/bin/env python3
import argparse
import polars as pl  # preferred over pandas

def main():
    # parse args, do work, write output

if __name__ == "__main__":
    main()

Key Rules

• Library priority: polars (as pl) > pandas (as pd) — use polars unless blocked
• CLI: argparse only — no click, typer, or fire
• Snakemake access: snakemake.input, snakemake.output, snakemake.params, snakemake.config
• Logging: print() — no logging module
• Paths: pathlib.Path preferred
• Shebang: #!/usr/bin/env python3 on ALL standalone scripts
• Error handling: raise ValueError(...) for data issues, sys.exit(1) for CLI errors
• Output format: TSV (tab-separated) — .tsv extension
• Types: Optional type annotations on shared functions; scripts skip them

Script Categories

• Aggregation (aggregate_*.py): Read per-sample native quantifier outputs → produce unified count/TPM matrices
• Namespace (build_*.py): Construct tx2gene, gene namespace, StringTie bridge, import manifests
• Utilities (generate_samples.py, normalize_featurecounts_gtf.py, create_gene_mapping.py, summarize_microbe_clues.py)

R CONVENTIONS

Boilerplate

box::use(dplyr[...], readr[...], ...)
sink(snakemake@log)

# Read inputs from snakemake@input
data <- readr::read_tsv(snakemake@input[["matrix"]])

# ... analysis ...

# Write outputs to snakemake@output
readr::write_tsv(result, snakemake@output[["results"]])
sink()

Key Rules

• Package imports: box::use() for explicit namespace — dplyr, readr, DESeq2, ggplot2, tximport
• Snakemake access: snakemake@input, snakemake@output, snakemake@params, snakemake@config, snakemake@log, snakemake@threads
• Logging: sink(snakemake@log) at top — redirects all output
• DEA: DESeq2 only — no edgeR, no limma-voom
• Consensus: RobustRankAggreg for RRA, custom CCT (Cauchy Combination Test) — no dynamic weighting
• Lint suppression: # nolint start: object_usage_linter. / # nolint end for known false positives
• Helper operators: %||% (null coalesce), as_flag() (boolean parse)

ANTI-PATTERNS

• Don't add __init__.py — scripts are standalone, not a package
• Don't import between scripts — each runs independently per Snakemake rule
• Don't add click/typer/fire — argparse only for Python
• Don't use edgeR or limma in R — DEA is DESeq2-fixed
• Don't add new consensus stat methods without updating docs/methods.md
• Don't use logging module — print() only