Amino acid sequence assignment from single molecule peptide sequencing data using a two-stage classifier.

Area of Science:

• Proteomics
• Bioinformatics
• Machine Learning

Background:

• Fluorosequencing is a novel proteomics technology for single molecule protein sequencing.
• This method generates sparse amino acid sequence data for individual peptides.
• Analyzing this data requires advanced computational frameworks.

Purpose of the Study:

• To develop a machine learning-based interpretive framework, Whatprot, for analyzing fluorosequencing data.
• To improve the identification of peptides and parent proteins from complex mixtures.
• To enable more accurate protein sequencing error rate estimates.

Main Methods:

• Utilized Hidden Markov Models (HMMs) to model peptide states during fluorosequencing.
• Implemented a Bayesian classifier with HMMs.
• Incorporated a k-Nearest Neighbors (kNN) classifier for pre-filtering.
• Combined kNN and HMM classifiers in a hybrid approach.

Main Results:

• The hybrid kNN-HMM framework achieved tractable runtimes.
• Demonstrated acceptable precision and recall for peptide and protein identification.
• Outperformed individual kNN or HMM classifiers in performance.
• Enabled efficient interpretation of fluorosequencing data against a proteome reference database.

Conclusions:

• The Whatprot framework offers an efficient method for interpreting single molecule protein sequencing data.
• The hybrid kNN-HMM approach balances computational efficiency with high accuracy.
• This work facilitates the advancement of fluorosequencing technology and proteomic analysis.