National and Kapodistrian University of Athens
Efficient algorithms and architectures for protein 3-D structure comparison
Abstract
dc:descriptionProtein Structure Comparison (PSC) is a well developed field of computational proteomics with active interest since it is widely used in structural biology and drug discovery. Fast increasing computational demand for all-to-all protein structures comparison is a result of mainly three factors: rapidly expanding structural proteomics databases, high computational complexity of pairwise PSC algorithms, and the trend towards using multiple criteria for comparison and combining their results (MCPSC). Despite the sustained interest in the field over the past three decades there are still open challenges in large-scale MCPSC. Firstly, its application using modern many-core and multi-core processor architectures remains unexplored. Secondly, there are few works that apply MCPSC to develop biologically relevant clusters and classify proteins. Finally, there is lack of bioinformatics software tools to support comparative analysis of large protein datasets on commodity computers.In order to address these challenges, in this thesis we have developed a software framework that exploits many-core and multi-core CPUs to implement efficient parallel MCPSC schemes in modern processors based on three popular PSC methods, namely, TMalign, CE, and USM. We evaluate and compare the performance and efficiency of two parallel MCPSC implementations using Intel’s experimental many-core Single-Chip Cloud Computer (SCC) CPU as well as Intel’s Core i7 multi-core processor. Further, we have developed a Python based utility, called pyMCPSC, allowing users to perform MCPSC efficiently, by exploiting the parallelism afforded by the multi-core CPUs of today’s desktop computers. We show how pyMCPSC, which combines five PSC methods and five different consensus scoring schemes, facilitates the analysis of similarities in protein domain datasets and how it can be easily extended to incorporate more PSC methods in the consensus scoring as they are becoming available.Experimental results from our analysis, show that the 48-core Intel SCC NoC processor is more efficient than the latest generation Core i7 CPU, achieving a speedup factor of 42 (efficiency of 0.9), making many-core processors an exciting technology for large-scale structural proteomics. We compare and contrast the performance of the two processors on several benchmark datasets and also show that MCPSC outperforms its component PSC methods in grouping related domains, achieving a high F-measure of 0.91 on the CK34 dataset. We further demonstrate using the Proteus300 dataset, that consensus MCPSC scores form a reliable basis for identifying the true classification of a protein domain, as evidenced both by ROC analysis as well as Nearest-Neighbor analysis. Structure similarity based “Phylogenetic Trees” generated by consensus scores provide insight into functional grouping within the dataset of domains. Furthermore, scatter plots generated by pyMCPSC reveal the existence of strong correlation between protein domains belonging to SCOP Class C and loose correlation between those of SCOP Class D within the Proteus dataset. Such analyses and corresponding visualizations help users quickly gain insights about their datasets. Finally we demonstrate that even very large datasets (such as SCOPCATH) can be processed and consensus scores based structural analysis carried out on readily available multi-core processors using our developed methods. pyMCPSC has been released to the proteomics community through GitHub and can be accessed at https://github.com/xulesc/pymcpsc.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Sharma, Anuj
- Ανούτζ, Σάρμα
Subjects
dc:subject × 14Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44985