2019/09/04 by John T. Halloran, David M. Rocke, Halloran, John T. +1
Computer Science · Engineering · #Bayesian Modeling and Causal Inference #FOS: Biological sciences #FOS: Computer and information sciences #Fault Detection and Control Systems #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Data Classification #Neural Networks and Applications #Quantitative Methods (q-bio.QM)
paper · pdf · doi:10.48550/arxiv.1909.02093
openalex publication_date 2019/09/04 · openalex created_date 2022/09/13 · openalex updated_date 2026/07/28
Tandem mass spectrometry (MS/MS) is a high-throughput technology used\ntoidentify the proteins in a complex biological sample, such as a drop of\nblood. A collection of spectra is generated at the output of the process, each\nspectrum of which is representative of a peptide (protein subsequence) present\nin the original complex sample. In this work, we leverage the log-likelihood\ngradients of generative models to improve the identification of such spectra.\nIn particular, we show that the gradient of a recently proposed dynamic\nBayesian network (DBN) may be naturally employed by a kernel-based\ndiscriminative classifier. The resulting Fisher kernel substantially improves\nupon recent attempts to combine generative and discriminative models for\npost-processing analysis, outperforming all other methods on the evaluated\ndatasets. We extend the improved accuracy offered by the Fisher kernel\nframework to other search algorithms by introducing Theseus, a DBN representing\na large number of widely used MS/MS scoring functions. Furthermore, with\ngradient ascent and max-product inference at hand, we use Theseus to learn\nmodel parameters without any supervision.\n