How much storage precision can be lost: Guidance for near-lossless compression of untargeted metabolomics mass spectrometry data

Abstract

The size of high-resolution mass spectrometry (HRMS) data has been increasing significantly. Several lossy compressors have been developed for higher compression rates. Currently, a comprehensive evaluation of what and how MS data (m/z and intensities) with precision losses would affect data processing (i.e., feature detection and compound identification) is absent. Here, we set an error threshold at 1% to assess the significance of the difference between two files in feature and compound detection results obtained from MZmine3. First, we examined that mzML files with both m/z and intensity encoded in 32-bit precision appear to be a preferred combination via msConvert, which has smaller file size and minor variation with other combinations of storage precision (<0.13%). We then identified that the absolute error of 10-4 for m/z had a feature detection error of 0.57% and compound detection error of 1.1%. For intensities, the relative error group of 2x10-2 had an error of 4.65% for features and 0.98% for compounds, compared with precision-lossless files. Taken together, we provided a reasonable scene-accuracy proposal, with a maximum absolute error of 10-4 for m/z and a maximum relative error of 2x10-2 for intensity. This guidance aimed to help researchers in improving lossy compression algorithms and minimizing the negative effects of precision losses on downstream data processing.