add diffexp module

related to #39

CanDI/pipelines/diffexp.py

@@ -0,0 +1,52 @@
+import numpy as np
+import pandas as pd
+import anndata as ad
+
+from pydeseq2.dds import DeseqDataSet
+from pydeseq2.default_inference import DefaultInference
+from pydeseq2.ds import DeseqStats
+from adpbulk import ADPBulk
+
+
+def pseudobulk_by_group(adt, groups, method="mean"):
+    # initialize the object
+    adpb = ADPBulk(adt, groupby=groups, method=method)
+
+    # perform the pseudobulking
+    pseudobulk_matrix = adpb.fit_transform()
+
+    # retrieve the sample metadata (useful for easy incorporation with edgeR)
+    sample_meta = adpb.get_meta()
+
+    out = ad.AnnData(
+        X=pseudobulk_matrix,
+        obs=sample_meta.set_index('SampleName')
+    )
+
+    return out
+
+
+def run_deseq(adata, design, tested_level, ref_level, n_cpus=8):
+
+    inference = DefaultInference(n_cpus=n_cpus)
+
+    dds = DeseqDataSet(
+        counts=adata.to_df().astype(int),
+        metadata=adata.obs,
+        design_factors=design,  # compare samples based on the "condition"
+        refit_cooks=True,
+        inference=inference,
+    )
+
+    dds.deseq2()
+
+    stat_res = DeseqStats(
+        dds, 
+        contrast=[design, tested_level, ref_level], 
+        inference=inference
+    )
+    stat_res.summary()
+
+    df = stat_res.results_df
+
+    return df