Mixed-precision iterative refinement using tensor cores on GPUs to accelerate solution of linear systems.

Double-precision floating-point arithmetic (FP64) has been the de facto standard for engineering and scientific simulations for several decades. Problem complexity and the sheer volume of data coming from various instruments and sensors motivate researchers to mix and match various approaches to optimize compute resources, including different levels of floating-point precision. In recent years, machine learning has motivated hardware support for half-precision floating-point arithmetic. A primary challenge in high-performance computing is to leverage reduced-precision and mixed-precision hardware. We show how the FP16/FP32 Tensor Cores on NVIDIA GPUs can be exploited to accelerate the solution of linear systems of equations Ax = b without sacrificing numerical stability. The techniques we employ include multiprecision LU factorization, the preconditioned generalized minimal residual algorithm (GMRES), and scaling and auto-adaptive rounding to avoid overflow. We also show how to efficiently handle systems with multiple right-hand sides. On the NVIDIA Quadro GV100 (Volta) GPU, we achieve a 4 × - 5 × performance increase and 5× better energy efficiency versus the standard FP64 implementation while maintaining an FP64 level of numerical stability.

Novel approaches to visualization and data mining reveals diagnostic information in the low amplitude region of serum mass spectra from ovarian cancer patients.

The ability to identify patterns of diagnostic signatures in proteomic data generated by high throughput mass spectrometry (MS) based serum analysis has recently generated much excitement and interest from the scientific community. These data sets can be very large, with high-resolution MS instrumentation producing 1-2 million data points per sample. Approaches to analyze mass spectral data using unsupervised and supervised data mining operations would greatly benefit from tools that effectively allow for data reduction without losing important diagnostic information. In the past, investigators have proposed approaches where data reduction is performed by a priori "peak picking" and alignment/warping/smoothing components using rule-based signal-to-noise measurements. Unfortunately, while this type of system has been employed for gene microarray analysis, it is unclear whether it will be effective in the analysis of mass spectral data, which unlike microarray data, is comprised of continuous measurement operations. Moreover, it is unclear where true signal begins and noise ends. Therefore, we have developed an approach to MS data analysis using new types of data visualization and mining operations in which data reduction is accomplished by culling via the intensity of the peaks themselves instead of by location. Applying this new analysis method on a large study set of high resolution mass spectra from healthy and ovarian cancer patients, shows that all of the diagnostic information is contained within the very lowest amplitude regions of the mass spectra. This region can then be selected and studied to identify the exact location and amplitude of the diagnostic biomarkers.

Clinical proteomics and biomarker discovery.

Early detection of disease generally provides much-improved outcomes by a definitive medical procedure or through lifestyle modification along with specific medical management strategies. For serum biomarkers, which are central to the diagnosis of many diseases, to become truly useful sentinels of pathogenesis, their sensitivity and specificity in both early detection and recurrence monitoring must be improved. Currently, the detection and monitoring of disease markers is based on solitary proteins, and this approach is not always reliable. New classes of biomarkers derived from mass spectroscopy analysis of the low molecular weight proteome have shown improved abilities in the early detection of disease and hence in patient risk stratification and outcome. The development of a modular platform technology with sufficient flexibility and design abstractions allowing for concurrent experimentation, test, and refinement will help speed the progress of mass spectroscopy-derived proteomic pattern-based diagnostics from the scientific laboratory to the medical clinic. For acceptance by scientists, physicians, and regulatory personnel, new bioinformatic tools are essential system components for data management, analysis, and intuitive display of these new and complex data. Clinically engineered mass spectroscopy systems are essential for the further development and validation of multiplexed biomarkers that have shown tremendous promise for the early detection of disease.