We request funds to purchase a GPU-accelerated parallel computer to provide financially sustainable support for ongoing drug discovery efforts at Vanderbilt University. The proposed system consists of 20 Dell R720 compute nodes each equipped with dual nVidia K20 Tesla GPUs, 64GB of RAM, and dual 6-core Intel Xeon 2620 CPUs. The system will be made available to the Vanderbilt drug discovery research community through our Computational Structural and Chemical Biology Core (CSCBC), which offers a comprehensive menu of state-of-the-art molecular modeling techniques using either a self-service or fee-for-service consulting model. Expertise in these areas is provided by highly qualified staff scientists and PIs and in the Center for Structural Biology (CSB) and the Vanderbilt Institute for Chemical Biology (VICB). Fundamental to our approach is the recognition that successful application of computer modeling is dependent upon obtaining a sufficient quality and quantity of experimental data as inputs. The CSCBC therefore works closely, in a supporting role, with complimentary experimental cores such as the Crystallography, NMR, EPR, and Cryo-EM facilities sponsored by the CSB, and the High Throughput Screening and Chemical Synthesis cores sponsored by the VICB. For several years, GPUs have promised to dramatically accelerate software applications that can be organized to run on a simplified but highly parallel architecture, and do so at a relatively low cost when compared with traditional, more complex CPUs which are easier to program but also relatively more expensive per unit of raw computing power. Redesigning and reprogramming compute codes so that they can take advantage of GPU architectures remains an active area in computer science. However, many codes have begun to move out of a research and development phase into a production phase with broad adoption by end-users who are eager to leverage the price/performance advantage offered by GPUs. Virtual screening with quantitative structure activity relationships (QSAR) is one such mature application that has the potential to dramatically increase the number of active compounds that are identified in NIH-funded drug discovery projects at Vanderbilt. BCL:CHEMINFO is a GPU-accelerated ligand-based virtual screening application that can enrich the number of active hits in high-throughput screening assays by 30x-60x. BCL:CHEMINFO runs between 150x-300x faster on GPUs vs. CPUs, but it only costs about 2x to purchase GPU-equipped computers. Building the proposed GPU-accelerated computer to support drug discovery projects will significantly decrease the costs and time associated with discovering new molecules of therapeutic interest. The proposed system will be maintained at the Advanced Computing Center for Research and Education (ACCRE), which has a large 24/7 on-call staff with extensive experience managing clusters of CPU and GPU- based computers in support of biomedical research. The technical support at ACCRE will be largely covered by institutional matching funds, further reducing costs to the major and minor users on this proposal.
| Huang, Hui; Kuenze, Georg; Smith, Jarrod A et al. (2018) Mechanisms of KCNQ1 channel dysfunction in long QT syndrome involving voltage sensor domain mutations. Sci Adv 4:eaar2631 |
| Huo, Yuankai; Liu, Jiaqi; Xu, Zhoubing et al. (2018) Robust Multicontrast MRI Spleen Segmentation for Splenomegaly Using Multi-Atlas Segmentation. IEEE Trans Biomed Eng 65:336-343 |
| Huo, Yuankai; Xu, Zhoubing; Bao, Shunxing et al. (2018) Splenomegaly Segmentation using Global Convolutional Kernels and Conditional Generative Adversarial Networks. Proc SPIE Int Soc Opt Eng 10574: |
| Bobo, Meg F; Bao, Shunxing; Huo, Yuankai et al. (2018) Fully Convolutional Neural Networks Improve Abdominal Organ Segmentation. Proc SPIE Int Soc Opt Eng 10574: |
