An adaptive classification model for peptide identification.

Area of Science:

• Proteomics
• Mass Spectrometry
• Bioinformatics

Background:

• Accurate peptide sequence assignment is crucial for MS/MS-based protein identification.
• Existing post-database search algorithms struggle to differentiate ambiguous peptide spectrum matches (PSMs) from decoys.
• Target PSMs close to decoy PSMs are difficult to separate using discrimination functions alone.

Purpose of the Study:

• To develop a novel scoring system, CRanker, for improved peptide spectrum match (PSM) filtering.
• To assign confidence scores to target PSMs, enhancing the accuracy of protein identification.
• To address the challenge of separating true PSMs from false positives in large datasets.

Main Methods:

• A support vector machine (SVM)-based learning model was employed to determine optimal weights for target PSMs.
• A new scoring system, CRanker, was developed to rank target PSMs based on their confidence.
• Cholesky factorization was utilized to manage memory requirements for large PSM datasets by efficiently storing the kernel matrix.

Main Results:

• CRanker assigns a confidence weight to each target PSM, indicating its likelihood of being correct.
• The CRanker scoring system effectively ranks target PSMs, improving the reliability of protein identification.
• Cholesky factorization enabled efficient handling of large datasets by reducing memory usage.

Conclusions:

• CRanker identified more PSMs at similar false discovery rates compared to PeptideProphet and Percolator across various datasets.
• The proposed model demonstrated consistent performance across different test sets, validating its robustness.
• CRanker offers a more accurate and reliable method for peptide sequence assignment in MS/MS-based proteomics.