Highly accelerated feature detection in proteomics data sets using modern graphics processing units.

Motivation: Mass spectrometry (MS) is one of the most important techniques for high-throughput analysis in proteomics research. Due to the large number of different proteins and their post-translationally modified variants, the amount of data generated by a single wet-lab MS experiment can easily exceed several gigabytes. Hence, the time necessary to analyze and interpret the measured data is often significantly larger than the time spent on sample preparation and the wet-lab experiment itself. Since the automated analysis of this data is hampered by noise and baseline artifacts, more sophisticated computational techniques are required to handle the recorded mass spectra. Obviously, there is a clear tradeoff between performance and quality of the analysis, which is currently one of the most challenging problems in computational proteomics.

Results: Using modern graphics processing units (GPUs), we implemented a feature finding algorithm based on a hand-tailored adaptive wavelet transform that drastically reduces the computation time. A further speedup can be achieved exploiting the multi-core architecture of current computing devices, which leads to up to an approximately 200-fold speed-up in our computational experiments. In addition, we will demonstrate that several approximations necessary on the CPU to keep run times bearable, become obsolete on the GPU, yielding not only faster, but also improved results.

Availability: An open source implementation of the CUDA-based algorithm is available via the software framework OpenMS (http://www.openms.de).

Supplementary Information: Supplementary data are available at Bioinformatics online.