medsci-agent: Skill for OpenCode

PyDESeq2

Overview

PyDESeq2 is a Python implementation of the DESeq2 method for differential expression analysis of RNA-seq count data. It uses negative binomial generalized linear models with shrinkage estimation.

Typical Workflow

import pandas as pd
from pydeseq2.dds import DeseqDataSet
from pydeseq2.ds import DeseqStats

# counts: genes x samples DataFrame of raw counts (integers, unnormalized)
# metadata: samples DataFrame with condition column
counts = pd.read_csv("counts.csv", index_col=0)
metadata = pd.read_csv("metadata.csv", index_col=0)

# Create dataset
dds = DeseqDataSet(counts=counts, metadata=metadata, design="~condition")

# Run DESeq2 pipeline (size factors, dispersion, GLM fitting)
dds.deseq2()

# Statistical testing
stat_res = DeseqStats(dds, contrast=["condition", "treated", "control"])
stat_res.summary()

# Results DataFrame
results_df = stat_res.results_df
sig = results_df[results_df["padj"] < 0.05].sort_values("log2FoldChange")

Key Columns in Results

• baseMean: Mean normalized count across all samples.
• log2FoldChange: Effect size (positive = upregulated in numerator).
• pvalue: Raw p-value from Wald test.
• padj: Benjamini-Hochberg adjusted p-value.

Key Details

• Input must be raw (unnormalized) integer counts. Do NOT use TPM/FPKM.
• Counts matrix: rows = genes, columns = samples.
• Metadata index must match counts columns.
• contrast=["condition", "treated", "control"] means treated vs control.
• Apply LFC shrinkage with stat_res.lfc_shrink(coeff="condition_treated_vs_control").
• Filter low-count genes beforehand: keep genes with >= 10 counts in >= N samples.
• Install: pip install pydeseq2.