University of North Texas
Computational Methods for Peptide Identification and Protein Inference in Metaproteomics
Abstract
dc:descriptionMetaproteomics enables the functional profile of complex microbiomes in human, environmental and ecosystem contexts. However, accurate protein identification remains the computational bottleneck. Tandem mass spectra must be searched against vast, sample-specific protein sequences spanning hundreds of taxa; database incompleteness, sequence variation, post-translational modifications, and chimeric spectra degrade sensitivity and calibration. Cross-laboratory comparisons further reveal variability in identifications and false-discovery-rate (FDR) control, underscoring the need for robust rescoring, principled statistics, and transparent standards. Recent advances mitigate these issues through learning-based scoring and broader search strategies, but protein-level inference in metaproteomics remains intrinsically difficult because degenerate peptides map to homologous proteins across taxa, complicating grouping, taxonomic attribution, and protein-level FDR. This dissertation focuses on the protein identification pipeline—from mass spectra to confidently identified proteins. I contribute three methods that integrate machine learning with optimization. First, a deep learning–based peptide-spectrum-matching (PSM) rescoring model improves the sensitivity of the peptide-level at fixed FDR across heterogeneous microbiome datasets. Second, a curriculum-learning scoring framework exploits the structure in the PSM feature spaces to provide further gains while preserving practical throughput. Third, a species-aware protein inference formulation casts the degenerate-peptide problem as a linear optimization task that incorporates community composition as prior information, yielding taxonomically coherent protein groups with controlled protein-level error. Together, these contributions deliver end-to-end gains for metaproteomic identification by (i) increasing peptide-level confidence through data-driven rescoring and (ii) propagating those gains to protein-level calls via principled, prior-informed inference. The methods are evaluated on diverse datasets using decoy/entrapment controls and are released as open-source software to facilitate adoption in existing workflows.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Feng, Shichao
- Contributors dc:contributor
-
- Guo, Xuan
- Azad, Rajeev Kumar
- Zhao, Hui
- Yuan, Xiaohui
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Public
- Feng, Shichao 12/07/1992-
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_4655
ark: ark:/67531/metadc2566514