National Academy of Sciences
National Academy of Engineering
Institute of Medicine
National Research council
September 2, 2010
Jodi Swidzinski Hezky,
D. E. Shaw Research
120 West 45th Street, 39th Floor New York, NY 10036
Dear Dr. Hezky:
This letter details the work and transmits the final report of the Committee on Proposal Evaluation for Allocation of Supercomputing Time for the Study of Molecular Dynamics.
The committee evaluated submissions in response to a Request for Proposals (RFP) for Biomolecular Simulation Time on Anton, a specialized supercomputer designed and constructed by D.E. Shaw Research that allows for dramatically accelerated molecular dynamics simulations. D.E. Shaw is making time on a 512 node Anton machine available to the non-commercial research community without cost. During this time, the Anton machine will be housed at the Pittsburgh Supercomputing Center (PSC)’s National Resource for Biomedical Supercomputing; the related support work to enable Anton to be used by the community is supported by a grant from the National Institute of General Medical Sciences to the PSC. The work of the National Research Council (NRC) committee to evaluate proposals requesting allocations of time on Anton was supported by a contract between D.E. Shaw Research and the National Academy of Sciences and was performed under the auspices of the National Research Council’s Board on Life Sciences.
To undertake this task, the National Research Council convened a committee of experts to consider the proposals submitted in response to the Anton allocation RFP. The committee of 16 was chaired by Dr. Robert L. Jernigan, Director of the Baker Center for Bioinformatics and Biological Statistics and Professor of Biochemistry, Biophysics, and Molecular Biology at Iowa State University. The committee members were chosen for their expertise in molecular dynamics simulations, as well as in the subject areas represented in the 67 proposals that were considered by the committee for simulation time. The committee comprised a cross section of the biomolecular MD field, including both senior members and more junior investigators. The biographies of all of the committee members can be found in Appendix D.
The goal of the RFP for Biomolecular Simulation Time on Anton is to facilitate breakthrough science in the study of biomolecular systems by providing access to a dedicated, massively parallel supercomputer that allows significantly faster simulations of biomolecular systems using periodic boundary conditions and explicit solvent models. Anton’s capabilities allow questions to be addressed on multi-microsecond simulation timescales, so the program seeks to support
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 1
September 16, 2011 Jodi Swidzinski Hezky, Ph.D. D. E. Shaw Research 120 West 45th Street, 39th Floor New York, NY 10036 Dear Dr. Hezky: This letter describes the work and transmits the final report of the Committee on Proposal Evaluation for Allocation of Supercomputing Time for the Study of Molecular Dynamics, Second Round. The committee evaluated submissions received in response to a Request for Proposals (RFP) for Biomolecular Simulation Time on Anton, a supercomputer specially designed and built by D.E. Shaw Research (DESRES) that allows for dramatically increased molecular dynamics simulations compared to other currently available resources. Over the past year (October 1, 2010 – September 30, 2011), DESRES has made available to the non-commercial research community 3,000,000 node-hours on an Anton system housed at the Pittsburgh Supercomputing Center (PSC), based on the advice of a previous National Research Council committee convened in the fall of 2010. The success of the program has led DESRES to make the Anton machine housed at PSC available for an additional 3,000,000 node-hours over the 9-months following October 1, 2011 and DESRES has asked the National Research Council to, once again, facilitate the allocation of time to the non-commercial research community. The work of the National Research Council committee to evaluate proposals for time allocations was supported by a contract between D.E. Shaw Research and the National Academy of Sciences and was performed under the auspices of the National Research Council’s Board on Life Sciences. To undertake this task, the National Research Council convened a committee of experts to evaluate the proposals submitted in response to the aforementioned RFP. The committee of 17 was chaired by Dr. Robert L. Jernigan, Director of the Baker Center for Bioinformatics and Biological Sciences and Professor of Biochemistry, Biophysics and Molecular Biology at Iowa State University. The committee members were selected for their expertise in molecular dynamics simulations, as well as their experience in the subject areas represented in the 81 proposals that were considered by the committee. They comprised a cross section of the biomolecular dynamics field in academia, industry and government including an array of both senior and junior investigators. The committee was assisted by two external reviewers,1 who 1 The two external reviewers were Dr. Douglas Tobias, University of California, Irvine and Dr. Gerhard Hummer, National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health. Dr. Tobias submitted a proposal as a Principal Investigator for a time allocation on the Anton machine. Dr. Tobias provided the committee with an initial assessment of 8 proposals and provided his feedback to the committee only on those proposals. He took no part in other discussions or deliberations of the committee.
OCR for page 1
were selected to provide additional expertise in the areas of protein-lipid interactions and channel functions and who provided the committee with an initial assessment of a subset of the proposal submissions. The goal of the second RFP for Biomolecular Simulation Time on Anton has been to continue to facilitate breakthrough research in the study of biomolecular systems by providing a massively parallel system specially designed for molecular dynamics simulations. These special capabilities allow multi-microsecond to millisecond simulation timescales, which previously had been unobtainable. The program seeks to continue to support research that addresses important and high impact questions demonstrating a clear need for Anton’s special capabilities. The Anton RFP described the three criteria against which the committee was asked to evaluate proposals: Scientific Merit, including the potential to advance understanding on an important problem or question in the field; potential for breakthrough science resulting in new discoveries and understanding; the impact that successful completion of the proposed research would have on knowledge, methods, and current barriers in the field; and a scientifically and technologically feasible project with clear, well-developed, and appropriate goals, objectives, and approach to proposed studies. Justification for Requested Time Allocation, including a clear and well-justified need for multi-microsecond simulation timescales and a clear and convincing justification that the length and number of proposed simulation runs and node-hours requested are necessary and sufficient to achieve the scientific objectives. Investigator Qualifications and Past Accomplishments, including the appropriate experience and training to successfully conduct the proposed studies, evidence of knowledge and prior experience in molecular simulations, and past publications. Proposals from investigators who had previously received an allocation of time on Anton were required to include brief progress reports. Following guidance provided by DESRES and PSC, the committee did not use these progress reports as primary criteria but drew on them only to supplement its consideration of proposals. As explained in the RFP, staff at PSC conducted an initial assessment of all proposal submissions for completeness to determine whether they were technically feasible for simulation on Anton. A member of the PSC staff was also present as an observer throughout the review committee’s discussions to address technical specification that arose. In the second round of time allocations for Anton, DESRES and PSC will make time available at two levels. In the first level, approximately 15 proposals will receive an allocation of 100,000 node-hours each. The second level will include approximately 30 proposals to receive an allocation of 50,000 node-hours each. The committee was asked to identify proposals that best met the selection criteria defined above for allocations at each of the two levels. The committee was further asked to try to allocate approximately 50% of the time to investigators who did not receive an allocation in the first round. The judgments of the committee are based on which proposals adequately met or exceeded the selection criteria described above and on the estimates of required simulation time provided by the applicants. The committee was also permitted to Neither reviewer participated in the final committee discussion and assessment to achieve consensus on the list of proposals that best met the selection criteria. 2
OCR for page 1
consider a modified time allocation if it concluded that the proposed research required a greater or lesser number of node-hours than initially requested by an applicant. Initial reviews of the proposals were provided by the 17 committee members and the two external reviewers. Each proposal was assigned a minimum of two primary reviewers who were asked to evaluate the proposal based on the RFP and guidelines described above. Review assignments were made so that no proposal was evaluated by a reviewer from the applicant’s same institution or who had a collaborative relationship with an applicant. The NRC committee held its 2-day meeting in Washington, D.C. on August 5-6, 2011. On the first day of the meeting, members undertook a detailed discussion of the proposals. The two primary reviewers were asked to summarize their review for the committee, which was followed by discussion of the proposed research. As described in detail above, committee members considered the scientific merit, justification of the requested time and the qualifications of the principal investigator (PI). The discussion aimed at reaching consensus on which proposals best met the selection criteria. The committee divided the proposals into three groups. Group 1 contained proposals that best met or exceeded the selection criteria, Group 2 contained proposals that adequately met the selection criteria, and Group 3 contained proposals that met the selection criteria less well than those in Groups 1 and 2. If consensus could not be reached on an individual proposal on day 1, an additional committee member was assigned to review the proposal in detail for discussion on the second day. On the second day, committee members first discussed the proposals that were not easily categorized on day 1 and obtained further details from the additional reviewer, or a primary reviewer who had been unavailable on day 1. The committee then considered the slate of proposals as a whole, came to a consensus on the assignment of proposals into the three groups and, in some cases, decided to suggest a modified allocation of time on Anton. The committee concluded that the proposals listed below best meet the selection criteria set forth in the RFP for Biomolecular Simulation Time on Anton. Detailed comments for each of the 81 proposals are included in Appendix B. Group 1: The committee has identified 23 proposals that best met or exceeded the selection criteria. Of these 23 proposals, 5 proposals were selected for a modified allocation (identified below with an *). In numerical order by proposal submission number, these are: PSCA10006P Investigating protein folding and dynamics with a highly optimized additive force field; PI: Alexander MacKerell, University of Maryland, Baltimore [New user, identified for 50,000 node-hours]* PSCA10018P What are the atomistic underpinnings of molecular allostery and signaling?; PI: David Beratan, Duke University [New user, identified for 50,000 node- hours]* PSCA10030P Recognition of kinesin by microtubule during the stepping process; PI: Devarajan Thirumalai, University of Maryland, College Park [New user, identified for 100,000 node-hours] PSCA10031P Mapping slow dynamical regulation of a protein kinase by combining molecular dynamics with NMR data; PI: Susan Taylor, University of California, San Diego [New user, identified for 50,000 node-hours] 3
OCR for page 1
PSCA10038P Continued Exploration of the Human Adenovirus Protease Activation Pathway via Long Timescale Molecular Dynamics Simulations; PI: Ross Walker, University of California, San Diego [Returning user, identified for 100,000 node-hours] PSCA10039P Determining the pathway of nascent-protein insertion through the protein-conducting channel and into the membrane; PI: Klaus Schulten, University of Illinois, Urbana-Champaign [Returning user, identified for 100,000 node-hours] PSCA10040P Assessment of Multi-Microsecond Simulations of Intrinsically Disordered Proteins Using NMR: Applications to PDX1; PI: Scott Showalter, The Pennsylvania State University [Returning user, identified for 50,000 node-hours] PSCA10042P Determining Effects of HIV-1 gp41 Membrane-Spanning Domain on the Local Composition of a Mixed Cholesterol/Lipid Bilayer using Microsecond MD Simulation; PI: Cameron Abrams, Drexel University [New user, identified for 70,000 node-hours]* PSCA10043P Unraveling anomalous subdiffusion in heterogeneous membranes; PI: Edward Lyman, University of Delaware [New user, identified for 50,000 node-hours] PSCA10061P The determinants of C-type inactivation and recovery in the KcsA channel; PI: Benoit Roux, University of Chicago [Returning user, identified for 100,000 node-hours] PSCA10062P Using microsecond scale dynamics to characterize different classes of allosteric interactions; PI: J. Andrew McCammon, University of California, San Diego [Returning user, identified for 100,000 node-hours] PSCA10066P Molecular dynamics simulation of signal transduction in the squid rhodopsin G-protein coupled receptor; PI: Douglas Tobias, University of California, Irvine [Returning user, identified for 50,000 node-hours] PSCA10067P Dynamic coupling and fluctuations in protein-protein complexes; PI: Matthias Buck, Case Western Reserve University [Returning user, identified for 50,000 node-hours] PSCA10074P Chacterization of the structure and dynamics of a model two-domain protein using multi-microsecond simulations; PI: Lillian Chong, University of Pittsburgh [New user, identified for 50,000 node-hours] PSCA10076P Long time MD simulations to study large scale conformational transitions in RNA enzymes; PI: Darrin York, Rutgers University [New user, identified for 50,000 node-hours]* PSCA10081P Exploring Lipid-Protein Interactions Using Microsecond-scale Molecular Dynamics Simulation; PI: Toby Allen, University of California, Davis [Returning user, identified for 100,000 node-hours] PSCA10085P Propagation of conformational changes across a regulated catch-bond protein; PI: Wendy Thomas, University of Washington [New user, identified for 75,000 node-hours] 4
OCR for page 1
PSCA10086P Understanding the origin of high-fidelity co-translational protein targeting: Long-timescale simulations of the Signal Receptor Particle complex; PI: Thomas Miller, California Institute of Technology [Returning user, identified for 100,000 node-hours] PSCA10088P Entropy and Allostery in Protein-Ligand Binding via Simulations at the Microsecond Time Scale; PI: Michael Gilson, University of California San Diego [New user, identified for 100,000 node-hours] PSCA10095P Long Time Scale Molecular Dynamics Simulation of Protein Folding; PI: Martin Gruebele, University of Illinois, Urbana-Champaign [Returning user, identified for 100,000 node-hours] PSCA10096P Computational Design and Evaluation of Novel Enzyme Catalysts; PI: Kendell Houk, University of California, Los Angeles [Returning user, identified for 50,000 node-hours]* PSCA10097P The effects of nonnucleoside inhibitors on the structure and dynamics of HIV wild type reverse transcriptase and drug resistant mutants; PI: Michael Shirts, University of Virginia [New user, identified for 50,000 node-hours] PSCA10099P Folding of Ribosomal Signatures and Early tRNAs; PI: Zaida Luthey-Schulten, University of Illinois, Urbana-Champaign [New user, identified for 100,000 node-hours] Group 2: The committee has identified 21 proposals that adequately met the selection criteria. Of these 21 proposals, 11 proposals were selected for a modified allocation and are identified below with an *. In numerical order by proposal submission number, these are: PSCA10007P Molecular dynamics simulation study of the diffusion fence for PIP2 (phosphotidylinositol 4,5-bisphosphate) in lipid membrane; PI:Wonpil Im, The University of Kansas [New user, identified for 50,000 node-hours]* PSCA10008P Simulation of the two-step mechanism for kinesin force generation in realistic time scale: PI: Wonmuk Hwang, Texas A&M University [Returning user, identified for 100,000 node-hours] PSCA10014P Understanding the role of A-tracts in eukaryotic genome organization and their functions in transcriptional regulation; PI: George Schatz, Northwestern University [New user, identified for 45,000 node-hours] PSCA10017P Investigate Hidden Intermediates in Protein Folding; PI: Shuanghong Huo, Clark University [New user, identified for 40,000 node-hours]* PSCA10025P Influence of ATP and ADP on conformation and dynamics of the nucleotide-binding domain of Hsp70 chaperones; Harold Scheraga, Cornell University [New user, identified for 50,000 node-hours] 5
OCR for page 1
PSCA10029P Exploring the Membrane Selectivity and Toxicity of Antimicrobial Peptides; Judy Kim, University of California, San Diego [New user, identified for 60,000 node-hours] PSCA10052P Unraveling the Structure – Dynamics – Function Relationship of Human Histone Deacetylase 8; Peter Coveney, Yale University [Returning user, identified for 50,000 node-hours]* PSCA10056P Simulation and Analysis of Pressure Perturbation Dynamics of Proteins; PI: Kim Sharp, University of Pennsylvania [New user, identified for 50,000 node-hours] PSCA10059P In-silico assembly of the Urel urea channel from H. pylori from helical fragments; PI: Hartmut Luecke, University of California, Irvine [New user, identified for 50,000 node-hours] PSCA10060P Sequencing DNA Using MspA; PI: Aleksei Aksimentiev, University of Illinois, Urbana-Champaign [Returning user, identified for 50,000 node-hours]* PSCA10063P Continuous Long-Time Dynamics of RNA Molecules: Watching without Blinking for Microseconds through Anton’s Microscope: Extension Request; PI: Ioan Andricioaei, University of California, Irvine [Returning user, identified for 50,000 node- hours]* PSCA10065P Exploring the Gating Motions of Connexin Channels on a Microsecond Timescale; PI: Ivaylo Ivanov, Georgia State University [New user, identified for 50,000 node-hours] PSCA10068P Timescale of dynamics in unfolded/unstructured proteins: microseconds or nanoseconds?; PI: Jeetain Mittal, Lehigh University [New user, identified for 50,000 node-hours]* PSCA10070P Simple, Regulated Ion Channels: New Avenues for Synthetic Biology; PI: Andrew Pohorille, University of California, San Francisco [Returning user, identified for 50,000 node-hours] PSCA10071P Molecular Dynamics Simulations of Conformational Dynamics in the p38α MAP Kinase: Effects of the Apo State’s Flexibility on Inhibitor Binding Affinity; PI: Adrian Elcock, University of Iowa [Returning user, identified for 50,000 node-hours]* PSCA10073P Probing allosteric regulation in the Imidazole Glycerol Phosphate Synthase by microsecond MD simulations; PI: Victor Batista, Yale University [New user, identified for 50,000 node-hours]* PSCA10080P Anton Simulations of EPAC1 to illuminate the molecular details of cAMP based allosteric control of conformational change; PI: Thomas Woolf, Johns Hopkins University [New user, identified for 50,000 node-hours]* PSCA10090P Metabolite permeation and voltage-gating of the mitochondrial channel VDAC; PI: Michael Grabe, University of Pittsburgh [Returning user, identified for 50,000 node-hours]* 6
OCR for page 1
PSCA10093P Atomistic modeling of the resting and activated states of a voltage-gated potassium channel voltage-sensing domain; PI: J. Alfredo Freites, University of California, Irvine [Returning user, identified for 50,000 node-hours]* PSCA10100P Dynamics of Calcium-Dependent Processes Essential for Cadherin Function in Hearing and Deafness; PI: David Corey, HHMI & Harvard Medical School [Returning user, identified for 100,000 node-hours] PSCA10102P Capturing Large-Scale Structural Transitions in Membrane Transporters at Atomic Resolution; PI: Emad Tajkhorshid, University of Illinois, Urbana-Champaign [Returning user, identified for 100,000 node-hours] The time allocations for the 44 proposals identified by the committee as meeting or exceeding the selection criteria for time allocations total approximately 2,890,000 node-hours. Of the 44 proposals identified, 15 were identified at the 100,000 node-hour level and 29 at the 50,000 node- hour level.2 A total of 1,340,000 node-hours were allocated to 23 proposals whose principal investigator did not receive time on Anton during the past year (identified as “new users”). The remaining 1,550,000 node-hours are allocated to 21 proposals from investigators who had received first round time allocations (indentified as “returning users”). After considerable deliberation, the committee has concluded that the most appropriate use of the remaining 110,000 node-hours would be for the Pittsburgh Supercomputing Center to use its discretion to allocate additional time to Group 1 proposals under the following guidelines: 1. No PI should be allocated more than a total of 100,000 node-hours 2. PIs may petition PSC for additional time after they have used their initial allocation 3. Preference should be given to PIs who did not receive allocations in the first round In carrying out its task, the committee identified as many promising proposals as possible given the constraints on the total available simulation time. The committee also encourages D.E. Shaw Research and the Pittsburgh Supercomputing Center to establish a collective repository to share data generated, because the trajectories obtained may be of use to other investigators in the community. The committee would like to thank D.E. Shaw Research, the Pittsburgh Supercomputing Center, and all of the 2011 Anton applicants for the opportunity to assist in identifying the proposals best meeting the selection criteria for time allocations on the Anton machine. The committee members were universally enthusiastic about the potential advances in the field that are facilitated by Anton and are looking forward to seeing the important new results from the Anton users. Sincerely, Robert L. Jernigan Chair cc: Dr. Markus Dittrich, Pittsburgh Supercomputing Center Dr. Warren Muir, National Research Council Dr. Frances Sharples, National Research Council 2 The 100,000 node-hour level is defined as proposals that were identified for 70,000 node-hours or greater. The 50,000 node-hour level is defined as proposals that were identified for less than 70,000 node-hours. 7
OCR for page 1
Appendices: A. Table 1: Proposals Reviewed by the Committee B. Individual Proposal Summary Evaluations C. Proposal Evaluation Criteria D. Roster and Biographical Sketches of Committee Members and Reviewers E. The Board on Life Sciences, the Board on Chemical Sciences and Technology, and the National Academies F. Acknowledgment of Report Reviewer 8
OCR for page 1