Fix mle ValueError on count tables with spaces in sgRNA/gene names (#7)

[davidliwei]

Fixes #7.

Problem

mageck2 mle crashes with a cryptic numpy error on a count table that mageck2 test (RRA) parses fine:

ValueError: setting an array element with a sequence. The requested array
has an inhomogeneous shape after 1 dimensions. The detected shape was (3,) + inhomogeneous part.

from np.matrix(ginst.nb_count) in read_gene_from_file.

Root cause

read_gene_from_file split count-table lines on generic whitespace (line.strip().split()), while the rest of mageck2 splits on tab (e.g. mageckCount.py, splitter='\t').

When an sgRNA or gene name contains a space — common for control entries such as Non-Targeting Control in a with-control library — the whitespace split shifts the count columns. A non-numeric token (Control) is then read as a count; float() raised ValueError, which was silently swallowed without appending. That left one sample's list one element short, making nb_count ragged and crashing later in np.matrix(). RRA was unaffected because it splits on tab.

Reproduced with a minimal table containing a spaced gene name; the real read_gene_from_file now parses it correctly.

Changes

• Split non-CSV count tables on \t, consistent with the rest of mageck2 (the actual fix).
• Fail loudly with line/column/value context on an unparseable count, instead of silently producing a ragged matrix and an opaque downstream crash.

🤖 Generated with Claude Code