Score-direction inference has no override and silently defaults on ties

[ArthurDeclercq]

ms2rescore/parse_psms.py:210

lower_score_is_better = n_lower_better > n_higher_better returns False on a tie (or weak signal / few decoys). For a genuinely lower-is-better engine this inverts set_ranks, so the rank filter (:47-49) keeps the worst candidates and drops the best. The lower_score_is_better config option was removed, so the user cannot correct a misinference. Consider restoring an explicit override.

[RalfG]

I consider this a non-issue. If there's a tie, it means that in either direction, the score is not doing a good job distinguishing good from bad targets. With either option, we'll have a similar amount of low-quality PSMs for calibrating, leading to worse feature generation. Garbage in == garbage out, even for ms2rescore 😛