MOTIVATION: Protein-protein interactions (PPI) play important roles in cellular activities. Due to the technical difficulty and high cost of experimental methods, there are considerable interests towards the development of computational approaches, such as protein docking, to decipher PPI patterns. One of the important and difficult aspects in protein docking is recognizing near-native conformations from a set of decoys, but unfortunately, traditional scoring functions still suffer from limited accuracy. Therefore, new scoring methods are pressingly needed in methodological and/or practical implications. RESULTS: We present a new deep learning-based scoring method for ranking protein-protein docking models based on a 3D RepVGG network, named TRScore. To recognize near-native conformations from a set of decoys, TRScore voxelizes the protein-protein interface into a 3D grid labeled by the number of atoms in different physicochemical classes. Benefiting from the deep convolutional RepVGG architecture, TRScore can effectively capture the subtle differences between energetically favorable near-native models and unfavorable non-native decoys without needing extra information. TRScore was extensively evaluated on diverse test sets including protein-protein docking benchmark 5.0 update set, DockGround decoy set, as well as realistic CAPRI decoy set and overall obtained a significant improvement over existing methods in cross-validation and independent evaluations. AVAILABILITY AND IMPLEMENTATION: Codes available at: https://github.com/BioinformaticsCSU/TRScore.

Cryo-electron microscopy (cryo-EM) has become one of important experimental methods in structure determination. However, despite the rapid growth in the number of deposited cryo-EM maps motivated by advances in microscopy instruments and image processing algorithms, building accurate structure models for cryo-EM maps remains a challenge. Protein secondary structure information, which can be extracted from EM maps, is beneficial for cryo-EM structure modeling. Here, we present a novel secondary structure annotation framework for cryo-EM maps at both intermediate and high resolutions, named EMNUSS. EMNUSS adopts a three-dimensional (3D) nested U-net architecture to assign secondary structures for EM maps. Tested on three diverse datasets including simulated maps, middle resolution experimental maps, and high-resolution experimental maps, EMNUSS demonstrated its accuracy and robustness in identifying the secondary structures for cyro-EM maps of various resolutions. The EMNUSS program is freely available at http://huanglab.phys.hust.edu.cn/EMNUSS.

MOTIVATION: Advances in microscopy instruments and image processing algorithms have led to an increasing number of Cryo-electron microscopy (cryo-EM) maps. However, building accurate models for the EM maps at 3-5 Å resolution remains a challenging and time-consuming process. With the rapid growth of deposited EM maps, there is an increasing gap between the maps and reconstructed/modeled three-dimensional (3D) structures. Therefore, automatic reconstruction of atomic-accuracy full-atom structures from EM maps is pressingly needed. RESULTS: We present a semi-automatic de novo structure determination method using a deep learning-based framework, named as DeepMM, which builds atomic-accuracy all-atom models from cryo-EM maps at near-atomic resolution. In our method, the main-chain and Cα positions as well as their amino acid and secondary structure types are predicted in the EM map using Densely Connected Convolutional Networks. DeepMM was extensively validated on 40 simulated maps at 5 Å resolution and 30 experimental maps at 2.6-4.8 Å resolution as well as an Electron Microscopy Data Bank-wide dataset of 2931 experimental maps at 2.6-4.9 Å resolution, and compared with state-of-the-art algorithms including RosettaES, MAINMAST and Phenix. Overall, our DeepMM algorithm obtained a significant improvement over existing methods in terms of both accuracy and coverage in building full-length protein structures on all test sets, demonstrating the efficacy and general applicability of DeepMM. AVAILABILITY AND IMPLEMENTATION: http://huanglab.phys.hust.edu.cn/DeepMM. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Protein structure alignment is one of the fundamental problems in computational structure biology. A variety of algorithms have been developed to address this important issue in the past decade. However, due to their heuristic nature, current structure alignment methods may suffer from suboptimal alignment and/or over-fragmentation and thus lead to a biologically wrong alignment in some cases. To overcome these limitations, we have developed an accurate topology-independent and global structure alignment method through an FFT-based exhaustive search algorithm, which is referred to as FTAlign. RESULTS: Our FTAlign algorithm was extensively tested on six commonly used datasets and compared with seven state-of-the-art structure alignment approaches, TMalign, DeepAlign, Kpax, 3DCOMB, MICAN, SPalignNS and CLICK. It was shown that FTAlign outperformed the other methods in reproducing manually curated alignments and obtained a high success rate of 96.7 and 90.0% on two gold-standard benchmarks, MALIDUP and MALISAM, respectively. Moreover, FTAlign also achieved the overall best performance in terms of biologically meaningful structure overlap (SO) and TMscore on both the sequential alignment test sets including MALIDUP, MALISAM and 64 difficult cases from HOMSTRAD, and the non-sequential sets including MALIDUP-NS, MALISAM-NS, 199 topology-different cases, where FTAlign especially showed more advantage for non-sequential alignment. Despite its global search feature, FTAlign is also computationally efficient and can normally complete a pairwise alignment within one second. AVAILABILITY AND IMPLEMENTATION: http://huanglab.phys.hust.edu.cn/ftalign/.

MOTIVATION: Given the importance of protein-ribonucleic acid (RNA) interactions in many biological processes, a variety of docking algorithms have been developed to predict the complex structure from individual protein and RNA partners in the past decade. However, due to the impact of molecular flexibility, the performance of current methods has hit a bottleneck in realistic unbound docking. Pushing the limit, we have proposed a protein-ensemble-RNA docking strategy to explicitly consider the protein flexibility in protein-RNA docking through an ensemble of multiple protein structures, which is referred to as MPRDock. Instead of taking conformations from MD simulations or experimental structures, we obtained the multiple structures of a protein by building models from its homologous templates in the Protein Data Bank (PDB). RESULTS: Our approach can not only avoid the reliability issue of structures from MD simulations but also circumvent the limited number of experimental structures for a target protein in the PDB. Tested on 68 unbound-bound and 18 unbound-unbound protein-RNA complexes, our MPRDock/DITScorePR considerably improved the docking performance and achieved a significantly higher success rate than single-protein rigid docking whether pseudo-unbound templates are included or not. Similar improvements were also observed when combining our ensemble docking strategy with other scoring functions. The present homology model-based ensemble docking approach will have a general application in molecular docking for other interactions. AVAILABILITY AND IMPLEMENTATION: http://huanglab.phys.hust.edu.cn/mprdock/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Summary: A structural database of peptide-protein interactions is important for drug discovery targeting peptide-mediated interactions. Although some peptide databases, especially for special types of peptides, have been developed, a comprehensive database of cleaned peptide-protein complex structures is still not available. Such cleaned structures are valuable for docking and scoring studies in structure-based drug design. Here, we have developed PepBDB-a curated Peptide Binding DataBase of biological complex structures from the Protein Data Bank (PDB). PepBDB presents not only cleaned structures but also extensive information about biological peptide-protein interactions, and allows users to search the database with a variety of options and interactively visualize the search results. Availability and implementation: PepBDB is available at http://huanglab.phys.hust.edu.cn/pepbdb/.

Basic data flow patterns that we call \textbf{performance idioms}, such as stream, transpose, reduction, random access and stencil, are common in scientific numerical applications. We hypothesize that a small number of idioms can cover most programming constructs that dominate the execution time of scientific codes and can be used to approximate the application performance. To check these hypotheses, we proposed an automatic idioms recognition method and implemented the method, based on the open source compiler Open64. With the NAS Parallel Benchmark (NPB) as a case study, the prototype system is about $90%$ accurate compared with idiom classification by a human expert. Our results showed that the above five idioms suffice to cover $100%$ of the six NPB codes (MG, CG, FT, BT, SP and LU). We also compared the performance of our idiom benchmarks with their corresponding instances in the NPB codes on two different platforms with different methods. The approximation accuracy is up to $96.6%$. The contribution is to show that a small set of idioms can cover more complex codes, that idioms can be recognized automatically, and that suitably defined idioms may approximate application performance.

Emerging storage technologies such as flash memories, phase-change memories, and spin-transfer torque memories are poised to close the enormous performance gap between disk-based storage and main memory. We evaluate several approaches to integrating these memories into computer systems by measuring their impact on IO-intensive, database, and memory-intensive applications. We explore several options for connecting solid-state storage to the host system and find that the memories deliver large gains in sequential and random access performance, but that different system organizations lead to different performance trade-offs. The memories provide substantial application-level gains as well, but overheads in the OS, file system, and application can limit performance. As a result, fully exploiting these memories' potential will require substantial changes to application and system software. Finally, paging to fast non-volatile memories is a viable option for some applications, providing an alternative to expensive, powerhungry DRAM for supporting scientific applications with large memory footprints.

Data intensive computing can be defined as computation involving large datasets and complicated I/O patterns. Data intensive computing is challenging because there is a five-orders-of-magnitude latency gap between main memory DRAM and spinning hard disks; the result is that an inordinate amount of time in data intensive computing is spent accessing data on disk. To address this problem we designed and built a prototype data intensive supercomputer named DASH that exploits flash-based Solid State Drive (SSD) technology and also virtually aggregated DRAM to fill the latency gap . DASH uses commodity parts including Intel® X25-E flash drives and distributed shared memory (DSM) software from ScaleMP®. The system is highly competitive with several commercial offerings by several metrics including achieved IOPS (input output operations per second), IOPS per dollar of system acquisition cost, IOPS per watt during operation, and IOPS per gigabyte (GB) of available storage. We present here an overview of the design of DASH, an analysis of its cost efficiency, then a detailed recipe for how we designed and tuned it for high data-performance, lastly show that running data-intensive scientific applications from graph theory, biology, and astronomy, we achieved as much as two orders-of- magnitude speedup compared to the same applications run on traditional architectures.

Basic data flow patterns which we call idioms, such as stream, transpose, reduction, random access and stencil, are common in scientific numerical applications. We hypothesize that a small number of idioms can cover most programming constructs that dominate the execution time of scientific codes and can be used to approximate the application performance. In this paper, we start with a manual analysis of code coverage on the NAS Parallel Benchmark (NPB) and find that five idioms suffice to cover 100% of the NPB codes. We then compare the performance of our idiom benchmarks and their corresponding instances in different NPB codes on two different platforms and find that they differ by about 30%. To check the hypotheses with real applications further, we propose an automatic idioms recognition method, implement the method basing on the open source compiler Open64, and verify the prototype system with the previous manual analysis results.