Non-descriptive IndexError when PEPs (or q-values?) cannot be calculated from score distribution

[RalfG]

Hi,

As part of MS²Rescore, which uses Mokapot, which uses Triqler for PEP calculation, we received an error report with a non-descriptive IndexError (compomics/ms2rescore#165 (comment)). This occurred because the input score distribution only contain 0 and 100. It therefore makes sense that PEP values could not be calculated. Perhaps a more descriptive error message could be produced?

Here's the relevant part of the stack trace:

File mokapot/confidence.py:476, in LinearConfidence._assign_confidence(self, desc)
    474 LOGGER.info("Assiging PEPs to %s...", level)
    475 try:
--> 476     _, pep = qvality.getQvaluesFromScores(
    477         scores[targets], scores[~targets], includeDecoys=True
    478     )
    479 except SystemExit as msg:
    480     if "no decoy hits available for PEP calculation" in str(msg):

File triqler/qvality.py:80, in getQvaluesFromScores(targetScores, decoyScores, includePEPs, includeDecoys, tdcInput, pi0, plotRegressionCurve, numBins)
     77 if plotRegressionCurve:
     78     scoresForPlot = evalScores.copy()
---> 80 probs = factor * np.exp(splineEval(evalScores, medians, variables))
     81 probs = monotonize(probs)
     83 if plotRegressionCurve:

File triqler/qvality.py:334, in splineEval(scores, medians, variables)
    325 derr = (g[1] - g[0]) / (medians[1] - medians[0]) - (
    326     medians[1] - medians[0]
    327 ) / 6 * gamma[0]
    328 scores[rights == 0] = (
    329     g[0] - (medians[0] - scores[rights == 0]) * derr
    330 )  # reuse "scores" array to save memory
    332 derl = (g[-1] - g[-2]) / (medians[-1] - medians[-2]) + (
    333     medians[-1] - medians[-2]
--> 334 ) / 6 * gamma[-3]
    335 scores[rights == n] = g[-1] + (scores[rights == n] - medians[-1]) * derl
    337 idxs = np.where((rights > 0) & (rights < n))

IndexError: index -3 is out of bounds for axis 0 with size 1

Thanks!
Ralf