« Back
PimMDN: Peptide Ion Mobility Mixture Density Network
Poster Title: PimMDN: Peptide Ion Mobility Mixture Density Network
Submitted on 05 Oct 2021
Author(s): Patrick Garrett, Robin Park, Titus Jung, Casimir Bamberger, John R. Yates, III
Affiliations: Department of Chemical Physiology, The Scripps Research Institute, La Jolla, CA, USA
Poster Views: 259
View poster »

Poster Information
Abstract: Due to the dynamic nature of an ionized peptide, a unique peptide sequence can adopt a range of possible mobilities 1 as well as multimodal or non parametric distributions. Standard machine learning methods, based on linear regression, are not suited to fit such distributions. They will learn to predict the mean of the distributions in order to minimize the mean squared error. Previous attempts to train deep learning models to predict a peptides ion mobility relied on pre processing the distribution into a single mean/median value. While these models demonstrate great capabilities, they lose valuable information regarding the effects that the peptide sequence has on its mobility distribution. Furthermore, these approaches cannot learn the effects which sequence pose on the standard deviation of the mobility distribution. Without a method to interpret the specificity, the true potential of mobility predictions cannot be utilized To combat these limitations, we created pimMDN which utilizes a Mixture Density Network (2) to model nonparametric and multimodal distributions, as a well as provide a metric of prediction specificity.Summary: Database searching and spectral library matching are the leading methods for quantifying and identifying peptides in mass spectrometry based proteomics. Though these methods can be improved through incorporating other peptide specific variables, such as ion mobility. We created PimMDN to maximize the contribution that ion mobility measurements can make to proteomic search methods. Current mobility models predict a single value but PimMDN predicts the entire mobility distribution.References: 1. Chih Hsiang Chang, Darien Yeung, Victor Spicer, Oleg Krokhin , Yasushi Ishihama bioRxiv 2020.09.14.296590;
2. Christopher M. Bishop, Mixture Density Networks (1994)
3. Meier F, Brunner AD, Koch S, Koch H, Lubeck M, Krause M, Goedecke N, Decker J, Kosinski T, Park MA, Bache N, Hoerning O, Cox J, Räther O, Mann M. Online parallel accumulation serial fragmentation (PASEF) with a novel trapped ion mobility mass spectrometer. Mol Cell Proteomics. 2018, PubMed: 30385480
4. Integrated Proteomics Pipeline (IP2) 1 for your publications
Report abuse »
Ask the author a question about this poster.
Ask a Question »

Creative Commons

Related Posters

Why Choose Untargeted Metabolomics?
Creative Proteomics

Lipidomics Research Methods
Creative Proteomics

Lipidomics and Bioinformatics Analysis
Creative Proteomics

Exosome Proteomics
Creative Proteomics

Untargeted Metabolomics
Creative Proteomics