Inspect

1. Preprocessing
   • Low intensity peaks are removed
     • Radius=25 Th, peak removed if not in top 6 for intensity
   • Parent mass correction
     • Move the parent mass by up to 2 Th to (till it matches precursor ion selected) to maximize b/y ion pairing
2. Tag Generation (MEAT)
   • De novo spectrum interpretation
     • Prefix Residue Mass spectra
     • Nodes, PRMs, tags…
     • Scoring
       • Intensity rank
       • PRM support: Empirical probability that it is a true peak (witness sets?)
       • Isotopic pattern matching
       • Additive scoring (combination of the 3 scores for each node through the path)
   • Implementation:
     • Tag length: 3
     • 50-100 top-scorers retained.
3. Database Search and Candidate Generation
   • Initial match
     • Search DB for tags by trie-based data structure http://en.wikipedia.org/wiki/Trie
   • Extension of initial match
     • Finding flanking sequences (allowing for modifications)
     • Decorations
       • Sorted list of potential mass adducts
     • Filters to remove impossible adducts
     • Dynamic programming assignment of PTM locations
     • Scorer adjusts for optimal attachment points for the PTMs.
   • Result constitutes a candidate peptide
4. Candidate Peptide Scoring
   • Scoring
     • Probabilistic model
       • Protein length bias correction
       • Bonuses for matching certain experimental criteria
   • P-value
     • 4 weighted scores are normalized and combined
     • Optimized weights are generated from training data sets (annotated T/F)