Skip to content

DeepLC/IM2Deep abort the entire run on any low-ID or tiny-calibration run #271

Description

@ArthurDeclercq

ms2rescore/feature_generators/deeplc.py:197; ms2rescore/feature_generators/im2deep.py:135

With the default calibration_set_size=None, the calibration count is (target_df["qvalue"] <= 0.01).sum(). In a multi-run dataset, a run whose targets are all above 1% global q-value yields 0 → head(0) empty → raise ValueError("Run 'X' has no target PSMs available for calibration.") even though target PSMs exist, and core.py's broad handler re-raises, killing the whole dataset. The DeepLC len == 0 guard is also insufficient for small non-empty sets: SplineTransformerCalibration.fit needs ≥10 PSMs (finder reports empirical crash for 1-9). The message is also misleading.

Fix: skip/degrade a single weak run instead of aborting; raise a clear, accurate message; guard the small-but-nonzero calibration case.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions