ポスター発表
- 第4日 5月18日（金） ポスター会場
-
4P-40（4C-O2-P-1605） PDF
Application of Non-negative Matrix Factorization in Mass Spectrometry-Based Shotgun Proteomics
Shotgun proteomics has been the core technology for profiling proteins, particularly in large-scale studies. The continuous advances in methodologies and mass spectrometry (MS) instrumentations generate a vast amount of high-quality information to be identified by subsequent computational platform. In spite of the need of high-performance tool for accurate protein identification, computational proteomics is bottlenecked by size and complexity in spectral processing. Non-negative matrix factorization (NMF) is a well-known machine learning technique in the field of signal processing for large-scale data processing. Currently, the existing NMF-based applications have increasingly reached out beyond problems in engineering to applications in molecular biology.
Here we present a novel application of NMF for protein identification from LC/MS data. While the proteomic profile is the result of mixtures of non-negative mass spectra and thus compatible to be processed by NMF, the signal derived from LC/MS-based shotgun proteomics experiment is uniquely complicated by noise, contamination, and wide dynamic range of protein intensities in the mixture. We illustrate the feasibility of the NMF algorithm to identify proteins from real experimental samples. We also propose the regularized algorithm by tailoring NMF to the unique characteristics of digested protein signals and benchmark the result with the conventional algorithm.