Seren Soner

Community-wide assessment of protein-interface modeling suggests improvements to design methodology

The CAPRI (Critical Assessment of Predicted Interactions) and CASP (Critical Assessment of protein Structure Prediction) experiments have demonstrated the power of community-wide tests of methodology in assessing the current state of the art and spurring progress in the very challenging areas of protein docking and structure prediction. We sought to bring the power of community-wide experiments to bear on a very challenging protein design problem that provides a complementary but equally fundamental test of current understanding of protein-binding thermodynamics. We have generated a number of designed protein-protein interfaces with very favorable computed binding energies but which do not appear to be formed in experiments, suggesting that there may be important physical chemistry missing in the energy calculations. A total of 28 research groups took up the challenge of determining what is missing: we provided structures of 87 designed complexes and 120 naturally occurring complexes and asked participants to identify energetic contributions and/or structural features that distinguish between the two sets. The community found that electrostatics and solvation terms partially distinguish the designs from the natural complexes, largely due to the nonpolar character of the designed interactions. Beyond this polarity difference, the community found that the designed binding surfaces were, on average, structurally less embedded in the designed monomers, suggesting that backbone conformational rigidity at the designed surface is important for realization of the designed function. These results can be used to improve computational design strategies, but there is still much to be learned; for example, one designed complex, which does form in experiments, was classified by all metrics as a nonbinder.

Hot Spots in a Network of Functional Sites

It is of significant interest to understand how proteins interact, which holds the key phenomenon in biological functions. Using dynamic fluctuations in high frequency modes, we show that the Gaussian Network Model (GNM) predicts hot spot residues with success rates ranging between S 8–58%, C 84–95%, P 5–19% and A 81–92% on unbound structures and S 8–51%, C 97–99%, P 14–50%, A 94–97% on complex structures for sensitivity, specificity, precision and accuracy, respectively. High specificity and accuracy rates with a single property on unbound protein structures suggest that hot spots are predefined in the dynamics of unbound structures and forming the binding core of interfaces, whereas the prediction of other functional residues with similar dynamic behavior explains the lower precision values. The latter is demonstrated with the case studies; ubiquitin, hen egg-white lysozyme and M2 proton channel. The dynamic fluctuations suggest a pseudo network of residues with high frequency fluctuations, which could be plausible for the mechanism of biological interactions and allosteric regulation.

Integer Programming Based Heterogeneous CPU-GPU Cluster Scheduler for SLURM Resource Manager

We present an integer programming based heterogeneous CPU-GPU cluster scheduler for the widely used SLURM resource manager. Our scheduler algorithm takes windows of jobs and solves an allocation problem in which free CPU cores and GPU cards are allocated collectively to jobs so as to maximize some objective function. We perform realistic SLURM emulation tests using the Effective System Performance (ESP) workloads. The test results show that our scheduler produces better resource utilization and shorter average job waiting times. The SLURM scheduler plug-in that implements our algorithm is available at http://code.google.com/p/slurm-ipsched/.

Integer programming based heterogeneous CPU-GPU cluster schedulers for SLURM resource manager

We present two integer programming based heterogeneous CPU-GPU cluster schedulers, called IPSCHED and AUCSCHED, for the widely used SLURM resource manager. Our scheduler algorithms take windows of jobs and solve allocation problems in which free CPU cores and GPU cards are allocated collectively to jobs so as to maximize some objective functions. Our AUCSCHED scheduler employs an auction based approach in which bids for contiguous blocks of resources are generated for each job. We perform realistic SLURM emulation tests using the Effective System Performance (ESP) and our own synthetic workloads. Even though it is difficult to generalize, the tests roughly show that out of the three scheduling plug-ins, AUCSCHED achieves better utilization, spread and packing, IPSCHED achieves better waiting time and SLURM Backfill achieves better fragmentation performances when compared with each other. The SLURM scheduler plug-ins that implement our algorithm are available at http://code.google.com/p/slurm-ipsched/. Two integer programming based heterogeneous CPU-GPU cluster scheduling algorithms are developed for supercomputers.SLURM scheduler plug-ins are developed and realistic emulation tests are carried out on large clusters.Tests show that our algorithms perform better in terms of utilization, spread and packing over SLURMs backfill plug-in.

Generating multibillion element unstructured meshes on distributed memory parallel machines

We present a parallel mesh generator called PMSH that is developed as a wrapper code around the open source sequential Netgen mesh generator. Parallelization of the mesh generator is carried out in five stages: (i) generation of a coarse volume mesh; (ii) partitioning of the coarse mesh; (iii) refinement of coarse surface mesh to produce fine surface submeshes; (iv) remeshing of each fine surface submesh to get a final fine mesh; (v) matching of partition boundary vertices followed by global vertex numbering. A new integer based barycentric coordinate method is developed for matching distributed partition boundary vertices. This method does not have precision related problems of floating point coordinate based vertex matching. Test results obtained on an SGI Altix ICE X system with 8192 cores confirm that our approach does indeed enable us to generate multibillion element meshes in a scalable way.

DynaFace: Discrimination between Obligatory and Non-obligatory Protein- Protein Interactions Based on the Complex's Dynamics

Protein-protein interfaces have been evolutionarily-designed to enable transduction between the interacting proteins. Thus, we hypothesize that analysis of the dynamics of the complex can reveal details about the nature of the interaction, and in particular whether it is obligatory, i.e., persists throughout the entire lifetime of the proteins, or not. Indeed, normal mode analysis, using the Gaussian network model, shows that for the most part obligatory and non-obligatory complexes differ in their decomposition into dynamic domains, i.e., the mobile elements of the protein complex. The dynamic domains of obligatory complexes often mix segments from the interacting chains, and the hinges between them do not overlap with the interface between the chains. In contrast, in non-obligatory complexes the interface often hinges between dynamic domains, held together through few anchor residues on one side of the interface that interact with their counterpart grooves in the other end. In automatic analysis, 117 of 139 obligatory (84.2%) and 203 of 246 non-obligatory (82.5%) complexes are correctly classified by our method: DynaFace. We further use DynaFace to predict obligatory and non-obligatory interactions among a set of 300 putative protein complexes. DynaFace is available at: http://safir.prc.boun.edu.tr/dynaface.

A new auction-based scheduler for heterogeneous systems with moldable generic resources support

Slurm resource management system is used on many TOP500 supercomputers. We present a new auction‐based heterogeneous cluster scheduler plug‐in called AUCSCHED2. AUCSCHED2 contributes two major enhancements: the first is the extension of Slurm to support generic resource moldability by specification of resource ranges. The generic resources include accelerators like graphics processing unit or Xeon Phi. The current version of Slurm supports specification of node ranges but not of generic resource ranges. Such a feature can be very useful to run‐time auto‐tuning applications and systems that can make use of variable number of generic resources. The second enhancement involves the implementation of a new integer programming formulation in AUCSCHED2 that drastically reduces the number of variables. This allows faster solution and larger number of bids to be generated. Slurm emulation results are presented for the heterogeneous 1408 node Tsubame supercomputer, which has 12 cores and three graphics processing units on each of its nodes. AUCSCHED2 is available at https://github.com/aucsched/aucsched2. Copyright

Poster: co-allocation based scheduling for parallel systems

State-of-the-art supercomputers are made up of multiple types of resources. User jobs also have wide spectrum of resource requirements. Hence, a supercomputer can be thought of as a collection of heterogeneous resources with heterogeneous usage requirements from the users. Schedulers for such systems are challenged by several issues like scalability, GPU, topology and energy awareness. We view each scheduling step as solving a co-allocation problem, i.e. the problem of allocating multiple resources simultaneously to jobs. Collection of jobs can be repeatedly taken from the front of the job queue (i.e. a window of jobs) and a co-allocation problem formulated as an (integer) linear program (ILP/LP) can be solved. ILP formulations for single-type and multiple instances, a CPU-GPU and generalized systems are provided. Co-allocation solver is applied to both the window of jobs and the backfilled jobs. Simulation results show effectiveness of our approaches when compared with pure first-come-first-served schedulers.