Learning from Decoys to Improve the Sensitivity and Specificity of Proteomics Database Search Results

摘要: The statistical validation of database search results is a complex issue in bottom-up proteomics. correct and incorrect peptide spectrum match (PSM) scores overlap significantly, making an accurate assessment true matches challenging. Since the complete separation between false hits practically never achieved, there need for better methods rescoring algorithms to improve upon primary results. Here we describe calibration False Discovery Rate (FDR) estimation through dynamic FDR calculation method, FlexiFDR, which increases both sensitivity specificity Modelling simple linear regression on decoy different charge states, method maximized number positives reduced negatives several standard datasets varying complexity (18-mix, 49-mix, 200-mix) few (E. coli Yeast) obtained from wide variety MS platforms. net positive gain spectral identifications was up 14.81% 6.2% respectively. approach applicable methodologies- separate as well concatenated search, high mass accuracy, semi-tryptic modification searches. FlexiFDR also applied Mascot showed performance than before. We have shown that appropriate threshold learnt decoys, can be very effective improving adapts itself instruments, data types It learns sets flexible automatically aligns underlying variables quality size.