NCSI: GPU Optimization of Protein Structure and Function Prediction Algorithms

Status	Completed
Mentor Name	Elijah MacCarthy
Mentor's XSEDE Affiliation	Campus Champion
Mentor Has Been in XSEDE Community	4-5 years
Project Title	GPU Optimization of Protein Structure and Function Prediction Algorithms
Summary	In this project, underrepresented and minority students at Lane College will be trained in high-performance computing techniques to be used in optimizing computational algorithms. The project will port sequential protein structure and function prediction applications to the GPU using OpenACC and with the Nvidia Visual Profiler, optimize the ported application for improved efficiency. The optimized application will be tested on a benchmark dataset to ensure it produces results with a comparable accuracy.
Job Description	1. Familiarizing with bioinformatics code to be ported and optimized on the GPU 2. Refactoring serial code to ensure accelerator compatibility 3. Analyzing code and profiling to identify hotspots for optimization using GNU/PGI profilers, gprof and pgprof. 4. After identifying hotspots, OpenACC and CUDA for optimizing applications on GPU will be used to port target regions to the GPU for optimization. 5. Nvidia Visual Profiler (NVVP) will be used to expose accelerator utilization and further optimization 6. Collecting benchmark dataset of protein sequences from the protein data bank (PDB) for testing the accuracy of the optimized code. 7. Summarize findings into at least one scientific paper for publication.
Computational Resources	The optimizations will be performed on XSEDE bridges and SDSC Comet resources through the Campus Champion allocations for Lane College. GNU, PGI, OpenACC and CUDA tools on these two XSEDE resources will be used in the optimization. Serial code is written specifically to target GNU resources, thus, XSEDE GNU tools will be used in the initial analysis and profiling of code on bridges. Subsequently the PGI compiler and CUDA will be used for porting to the GPU and for accelerator utilization analysis.
Contribution to Community	This project will firstly enable the students to become knowledgeable with regards to the XSEDE environment, clusters and parallel computing. Thus, producing researchers that will contribute to the XSEDE community. Secondly, this project optimizes a major application used by protein structure predictors, therefore, speeding up the protein structure prediction process for computational biologists.
Position Type	Learner
Training Plan	First 4 weeks of the program will be dedicated to providing students thorough training in programming and parallel computing. The first week of the program will involve training in Cpp. The second week will further the Cpp training and provide an introduction to parallel computing. The Third week will involve a detailed training on OpenMP programming (shared memory programing) with some summary of MPI programming (distributed memory programming). The final week of the 4 weeks will be devoted to systematic and meticulous training in OpenACC with hands on practice using different scientific codes. The OpenACC training will continue throughout the remaining 6 weeks of the program as students perform hands on optimization task on the bioinformatics code and prepare a manuscript for publication.
Student Prerequisites/Conditions/Qualifications	Basic programming skills in CPP/FORTRAN is a plus but not required.
Duration	Summer
Start Date	05/03/2021
End Date	07/12/2021