25 fix putative log calculation

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "pdex"
-version = "0.1.14"
+version = "0.1.15"
 description = "Parallel differential expression for single-cell perturbation sequencing"
 readme = "README.md"
 authors = [{ name = "noam teyssier", email = "noam.teyssier@arcinstitute.org" }]

src/pdex/_utils.py

@@ -1,6 +1,14 @@
 import anndata as ad
 import numpy as np
 
+# A heuristic to determine if the data is log-transformed
+# Checks if the mean cell umi count is greater than a certain threshold
+# If the the mean cell umi count is < UPPER_LIMIT_LOG, it is assumed that the data is log-transformed
+#
+# This limit is set to 15 (log-data with >15 average UMI counts would mean an
+# average UMI count of ($ e^{15} - 1 = 3.26M $ ) which is unlikely at this point)
+UPPER_LIMIT_LOG = 15
+
 
 def guess_is_log(adata: ad.AnnData, num_cells: int | float = 5e2) -> bool:
     """Make an *educated* guess whether the provided anndata is log-transformed.
@@ -15,9 +23,10 @@ def guess_is_log(adata: ad.AnnData, num_cells: int | float = 5e2) -> bool:
     mask = np.random.choice(adata.shape[0], size=num_cells, replace=False)
 
     # Sum the matrix across the selected cell subset
-    sums = adata[mask].X.sum(axis=1)
+    sums = adata[mask].X.sum(axis=1)  # type: ignore
 
-    # Extract the fractional components of the array
-    decimals, _ = np.modf(sums)
+    # Determine the mean cell umi count
+    mean_umi_count = np.mean(sums)
 
-    return np.any(decimals != 0.0)
+    # Return True if the mean cell umi count is less than the upper limit
+    return bool(mean_umi_count < UPPER_LIMIT_LOG)