Accelerated Structural Bioinformatics for Drug Discovery 5

Computer-aided design is one of the critical components of modern drug discovery. Drug development is routinely streamlined using computational approaches to improve hit identification and lead selection, to enhance bioavailability, and to reduce toxicity. In the last decade, a mounting body of genomic knowledge has been accumulated due to advancements in genome-sequencing technologies, presenting great opportunities for pharmaceutical research. However, new challenges also arose because processing this large volume of data demands unprecedented computing resources. On the other hand, the state-of-the-art heterogeneous systems currently deliver petaflops of peak performance to accelerate scientific discovery. In this chapter, we describe the development and benchmarking of a parallel version of eFindSite, a structural bioinformatics algorithm for the identification of drug-binding sites in proteins and molecular fingerprint-based virtual screening. Thorough code profiling reveals that structure alignment calculations in eFindSite consume approximately 90% of the wall-clock time. Parallelizing this portion of the code using pragma-based OpenMP enables the desired performance improvements, scaling well with the number of computing cores. Compared to a serial version, the parallel code runs 11.8 and 10.1 times faster on the processor and the coprocessor, respectively; when both resources are utilized simultaneously, the speedup is 17.6. By comparing the serial and parallel versions of eFindSite, we show the OpenMP implementation of structure alignments for many-core devices. With minimal modifications, a complex, hybrid C++/Fortran77 code was successfully ported to a heterogeneous architecture yielding significant speedups. This demonstrates howmodern drug discovery can be accelerated by parallel systems equipped with IntelXeon PhiTM coprocessors. In this chapter, we show solutions to challenges in moving this code to parallelism that are lessons with wide applicability. For instance, we tackle porting extensive use of thread-unsafe common blocks in the Fortran77 code using OpenMP to make thread-private copies. We also enlarged stack sizes to avoid segmentation fault errors [ulimit -s]. Serial and parallel versions of eFindSite are freely available; please see “For more information” at the end of this chapter.