Most of the protein function prediction methods used today operate at the protein-level, where functional labels are assigned to, and transferred between, full-length proteins based on overall similarities. However, majority of proteins are composed of multiple domains and function by specifically interacting with other proteins or molecules, thus many functional as- sociations should be limited to specific regions rather than the entire protein length. Existing domain-centric function prediction methods depend almost entirely on accurate domain fam- ily assignments to infer relationships between domains and functions, with regions that are unassigned to a known domain family left out of functional evaluation.
This work describes the development of an explicit region-specific function prediction methodology that localizes protein labels to pre-partitioned regions by enforcing overarching constraints to propagate labels between regions while preserving their existing assignments at the protein-level. We facilitate this by developing a text processing pipeline to integrate feature annotations from disparate data sources into a single feature-rich representation that more ad- equately encapsulates known functional features within regions, even in the absence of domain family assignments. The results were validated at the region-level using binding site annota- tions extracted from protein-ligand structures and show that the framework can successfully localize and improve protein-level prediction for site-specific functions. Our work reinforces the notion that different functions have different units of operation and should be treated as such in function prediction pipelines.
We extend this work by developing a deep learning approach to incorporate our dense and site-specific feature representation at the residue-level, without the need to first pre-partition sequences into discrete regions, using convolutional neural network models. We show that this approach can automatically localize function labels to regions and sites by applying visualization techniques used in image analysis to highlight functionally important residues and validating them across binding and catalytic functions associated with ligand-binding sites extracted from protein-ligand structures.
|Advisor:||Bonneau, Richard A.|
|Commitee:||Birnbaum, Kenneth, Coruzzi, Gloria, Ghedin, Elodie, Heguy, Adriana|
|School:||New York University|
|School Location:||United States -- New York|
|Source:||DAI-B 81/2(E), Dissertation Abstracts International|
|Keywords:||Computational biology, Data mining, Machine learning, Protein domains, Protein function prediction|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be