Shifted dot product scoring

shifted-dot-product-scoring

Summary

Shifted dot product scoring is a sensitive mass spectrum similarity metric that accounts for mass shifts caused by post-translational modifications by comparing query spectra against library spectra over a range of mass offsets. This enables detection of peptides bearing unknown or open modifications while maintaining strict false discovery rate control.

When to use

Apply this skill when matching query mass spectra to a spectral library in the presence of unknown post-translational modifications (PTMs) or non-enzymatic modifications. The query spectrum may represent a modified peptide whose unmodified counterpart exists in the library, but with m/z values shifted by the modification mass. Use this when you need to identify both unmodified and modified peptide-spectrum matches while controlling false discovery rate, particularly in open modification searches where the modification type is not pre-specified.

When NOT to use

• Input spectra are already annotated with specific known modifications: use standard dot product or cosine similarity scoring without mass shifting.
• Spectral library contains only unmodified peptides and you are searching for unmodified query spectra: standard spectral library search without cascade strategy is more efficient.
• Mass spectrometry data are low-resolution or have poor mass accuracy (>50 ppm): shifted dot product mass offsets require precise m/z values to distinguish genuine modifications from noise.

Inputs

• query spectra in mzML or mzXML format
• approximate nearest neighbor indexed spectral library
• mass offset range (Da) spanning expected modification masses
• candidate library spectrum set (pre-filtered by ANN indexing)

Outputs

• peptide-spectrum match (PSM) identifications with scores
• mass shift annotations for matched spectra
• false discovery rate filtered PSM results

How to apply

After approximate nearest neighbor indexing has selected a limited candidate set of library spectra, compute shifted dot product scores between the query spectrum and each candidate library spectrum across a range of mass offsets (typically spanning the modification mass search range). For each candidate, shift the query spectrum by a series of discrete mass deltas, compute the standard dot product (or cosine similarity) for each offset, and retain the maximum score across all tested shifts. This approach allows a single query to match library spectra with different mass shifts in a single pass. Apply false discovery rate filtering to the scored matches using a cascade strategy: first score unmodified matches, then iteratively score modified matches at higher mass shifts, resetting the FDR calculation at each cascade level to maximize sensitivity while maintaining strict error control.

Related tools

• ANN-SoLo (spectral library search engine that implements shifted dot product scoring within a cascade search strategy for open modification identification) — https://github.com/bittremieux/ANN-SoLo
• Faiss (approximate nearest neighbor indexing library underlying ANN-SoLo's candidate spectrum selection) — https://github.com/facebookresearch/faiss

Examples

# After indexing spectral library with ANN-SoLo, search query spectra and score with mass shifts:
# ann_solo search -i library.mgf -q query.mzML -o results.tsv -s 10 -e 200
# (implicitly applies shifted dot product scoring across mass offset range during cascade search)

Evaluation signals

• Verification that shifted dot product scores increase monotonically (or remain stable) as mass offset approaches the true modification mass, then decrease as offset moves away from true modification.
• Confirmation that identified modified peptides have consistent mass shifts matching known or expected PTM masses (phosphorylation +79.97 Da, acetylation +42.01 Da, etc.).
• False discovery rate for modified matches remains ≤ target threshold (typically 1% or 5%) when computed at each cascade level independently.
• Sensitivity (% of true positive PSMs identified) should be higher with shifted dot product scoring than with unshifted dot product when modifications are present, at equivalent false discovery rate.
• Query-library spectrum pairs with high shifted dot product scores should have visually similar peak patterns when the spectra are aligned at the identified mass offset.

Limitations

• Shifted dot product scoring increases computational cost linearly with the number of tested mass offsets; very wide modification mass search ranges (e.g., ±500 Da) may become prohibitively slow without GPU acceleration.
• The method assumes modification masses are known or can be bounded within a reasonable range; it will not detect modifications outside the pre-defined mass offset range.
• Performance degrades on low-resolution spectra with few peaks or high noise, as the dot product similarity metric becomes less discriminative.
• The cascade search strategy requires careful tuning of FDR thresholds at each level to balance sensitivity against false discovery; suboptimal thresholds can inflate error rates.

Evidence

• [readme] shifted dot product score to sensitively match modified spectra to their unmodified counterpart: "the shifted dot product score to sensitively match modified spectra to their unmodified counterpart"
• [intro] cascade search strategy to maximize identified unmodified and modified spectra while strictly controlling false discovery rate: "combined with a cascade search strategy to maximize the number of identified unmodified and modified spectra while strictly controlling the false discovery rate"
• [intro] approximate nearest neighbor indexing selects limited candidate spectra compared to unknown query spectrum: "ANN-SoLo uses approximate nearest neighbor indexing to speed up open modification searching by selecting only a limited number of the most relevant library spectra to compare to an unknown query"
• [intro] query spectra compared using shifted dot product scoring: "comparing each query to the selected candidate library spectra using shifted dot product scoring"