NMF-mGPU: non-negative matrix factorization on multi-GPU systems

Abstract

BackgroundIn the last few years, the Non-negative Matrix Factorization ( NMF ) technique has gained a great interest among the Bioinformatics community, since it is able to extract interpretable parts from high-dimensional datasets. However, the computing time required to process large data matrices may become impractical, even for a parallel application running on a multiprocessors cluster.In this paper, we present NMF-mGPU, an efficient and easy-to-use implementation of the NMF algorithm that takes advantage of the high computing performance delivered by Graphics-Processing Units ( GPUs ). Driven by the ever-growing demands from the video-games industry, graphics cards usually provided in PCs and laptops have evolved from simple graphics-drawing platforms into high-performance programmable systems that can be used as coprocessors for linear-algebra operations. However, these devices may have a limited amount of on-board memory, which is not considered by other NMF implementations on GPU.Results NMF-mGPU is based on CUDA ( Compute Unified Device Architecture ), the NVIDIA’s framework for GPU computing. On devices with low memory available, large input matrices are blockwise transferred from the system’s main memory to the GPU’s memory, and processed accordingly. In addition, NMF-mGPU has been explicitly optimized for the different CUDA architectures. Finally, platforms with multiple GPUs can be synchronized through MPI ( Message Passing Interface ). In a four-GPU system, this implementation is about 120 times faster than a single conventional processor, and more than four times faster than a single GPU device (i.e., a super-linear speedup).ConclusionsApplications of GPUs in Bioinformatics are getting more and more attention due to their outstanding performance when compared to traditional processors. In addition, their relatively low price represents a highly cost-effective alternative to conventional clusters. In life sciences, this results in an excellent opportunity to facilitate the daily work of bioinformaticians that are trying to extract biological meaning out of hundreds of gigabytes of experimental information. NMF-mGPU can be used “out of the box” by researchers with little or no expertise in GPU programming in a variety of platforms, such as PCs, laptops, or high-end GPU clusters. NMF-mGPU is freely available at https://github.com/bioinfo-cnb/bionmf-gpu.

Resolving single-cell heterogeneity from hundreds of thousands of cells through sequential hybrid clustering and NMF

“…We use a hybrid approach that combines accurate methods for cluster determination and cell classification, in combination with new algorithms for intelligent single-cell down-sampling. NMF has been shown to improve the detection of sub-populations from diverse datasets, due to its ability to identify interpretable parts from high dimensional datasets 32 . Using this refined workflow, we demonstrate improved performance over a large spectrum of existing approaches, across different datasets of varying complexity and size.…”

Deciphering programs of transcriptional regulation by combined deconvolution of multiple omics layers

“…If a GPU is available, an NMF implementation taking advantage of this highly parallel architecture for efficient matrix operations can be used by Bratwurst, e.g. NMF_GPU [24] (function runNmfGpu()). In the following, we present the details of a new implementation of an NMF-solver using the python modules PyCUDA -(ii) Iteration until convergence over update equations as used in [9].…”

“…NMF is a family of algorithms to factorize one large matrix V (of dimensions n x m) into two smaller matrices W (the signature matrix of dimensions n x k) and H (the exposure matrix of dimensions k x m) under the constraint of nonnegativity on all entries in both factor matrices W and H (Figure 1). k is called the factorization rank; a complexity reduction is achieved if k < n and k < m. In addition to a novel CUDA-based NMF implementation, the Bratwurst package allows to use different existing NMF implementations like the CUDA-based NMF_GPU [24] or a CPU implementation from the R package NMF [25]. The factorization rank k is a free parameter for any NMF method.…”

MPI-FAUN: An MPI-Based Framework for Alternating-Updating Nonnegative Matrix Factorization

“…Apart from distributed NMF algorithms using Hadoop and multicores, there are also implementations of the MU algorithm in a distributed memory setting using X10 [Grove et al 2014] and on a GPU [Mejía-Roa et al 2015].…”