In the current study, we show how ProCan90, a curated data set of HEK293 technical replicates, can be used to optimize the configuration options for algorithms in the OpenSWATH pipeline. Furthermore, we use this case study as a proof of concept for horizontal scaling of such a pipeline to allow 45 810 computational analysis runs of OpenSWATH to be completed within four and a half days on a budget of US $10 000. Through the use of Amazon Web Services (AWS), we have successfully processed each of the ProCan 90 files with 506 combinations of input parameters. In total, the project consumed more than 340 000 core hours of compute and generated in excess of 26 TB of data. Using the resulting data and a set of quantitative metrics, we show an analysis pathway that allows the calculation of two optimal parameter sets, one for a compute rich environment (where run time is not a constraint), and another for a compute poor environment (where run time is optimized). For the same input files and the compute rich parameter set, we show a 29.8% improvement in the number of quality protein (>2 peptide) identifications found compared to the current OpenSWATH defaults, with negligible adverse effects on quantification reproducibility or drop in identification confidence, and a median run time of 75 min (103% increase). For the compute poor parameter set, we find a 55% improvement in the run time from the default parameter set, at the expense of a 3.4% decrease in the number of quality protein identifications, and an intensity CV decrease from 14.0% to 13.7%.
Keywords: Amazon Web Services; HEK293; OpenSWATH; ProCan; big data; mass spectrometry; parameter optimization; proteomics; scalability; sensitivity analysis.