Repository navigation
ProteinAnalysis.protein_scale crashes for window 1 and silently biases even windows #5340
Description
ProteinAnalysis.protein_scale does not document an odd-window or minimum window restriction, but two valid-looking window choices seem to be mishandled:
window=1raisesZeroDivisionErrorwhile the natural one-residue profile is simply the selected scale value for each residue.- An even window double-counts
subsequence[window // 2]. The returned profile therefore depends on sequence direction even though all positional weights are described as symmetric about the window center.
Reproduction code:
from Bio.SeqUtils.ProtParam import ProteinAnalysis
scale = {aa: i + 1.0 for i, aa in enumerate("ACDEFGHIK")}
analysis = ProteinAnalysis("ACDEFGHIK")
analysis.protein_scale(scale, 1, edge=1.0)
# ZeroDivisionError: float division by zero
forward = analysis.protein_scale(scale, 4, edge=0.4)
reverse = ProteinAnalysis("KIHGFEDCA").protein_scale(
{aa: scale[aa] for aa in scale}, 4, edge=0.4
)[::-1]
print(max(abs(a - b) for a, b in zip(forward, reverse)))
# 0.2941176470588238
Activity
peterjc commentedon Oct 2, 2026
I'm guessing this was flagged with automated tools, rather than a biological use case?
Perhaps it would be simplest to document and impose limits on the window size.
ChenziqiAdam commentedon Oct 11, 2026
Sorry for the late reply. There is a real biological use case that hits it directly: the Hopp–Woods hydrophilicity scale
ProtParamData.hw, is defined for a hexapeptide window, i.e. an even window of 6.With the default
edge=1.0the profile should be the plain 6-residue average, but the middle residue is counted twice. The first two values are -0.543 and -0.686, where the plain mean gives -0.550 and -0.417. Reversing the sequence also changes the profile.So for this scale, an even window gives a silently wrong result rather than an error.
peterjc commentedon Oct 12, 2026
That's a good use-case then, thanks @ChenziqiAdam. Does the proposal on #5363 work for you?
ChenziqiAdam commentedon Oct 13, 2026
LGTM
Fix protein_scale for window 1 and even window sizes