How Much Storage Precision Can Be Lost: Guidance for Near-Lossless Compression of Untargeted Metabolomics Mass...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Bioinformatics

Background:

• Lossy compression offers superior compression rates for mass spectrometry (MS) data by reducing storage precision.
• The impact of precision loss on MS data processing remains unevaluated, hindering the development of effective lossy compressors.
• Evaluating precision loss is critical for optimizing MS data storage and analysis workflows.

Purpose of the Study:

• To evaluate the impact of different storage precisions (32-bit and 64-bit) on mass spectrometry data.
• To assess the effects of precision loss on MS data processing, including feature detection and compound annotation.
• To propose optimal precision error thresholds for lossy compression algorithms in MS data.

Main Methods:

• Evaluated 32-bit and 64-bit precision in lossless mzML files.
• Generated 10 precision-lossy files using truncation transformations (relative intensity errors and absolute m/z errors).
• Utilized MZmine3, XCMS for feature detection, and GNPS for compound annotation; compared Precision, Recall, F1-score, and file sizes.

Main Results:

• The discrepancy between 32-bit and 64-bit precision was less than 1%.
• An absolute m/z error of 10^-4 and relative intensity error of 2x10^-2 (5% error threshold) achieved F1-scores above 95%.
• For a stricter 1% error threshold (F1-scores >99%), absolute m/z error of 2x10^-5 and relative intensity error of 2x10^-3 are advised.

Conclusions:

• Lossy compression of MS data is feasible with minimal impact on downstream analysis when appropriate precision thresholds are applied.
• Recommended error thresholds balance compression efficiency with data integrity for feature detection and compound annotation.
• This study provides crucial guidance for developing improved lossy compression algorithms for mass spectrometry data.