Strategies to deal with batch effects with high resolution Orbitrap mass spectrometry for NOM characterisation

There is growing evidence that changes in the molecular composition of natural organic matter (NOM) in water drives changes in the effectiveness of water treatment processes. Hence, there is a growing interest in obtaining more detailed characterisation of natural organic matter, than traditional methods can provide. High resolution mass spectrometry is one such technique that is increasingly being used for NOM analysis. Predominately, Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS), has been used but Orbitrap MS is emerging as a more available, smaller, and cheaper alternative.  

Due to the high sensitivity of high-resolution mass spectrometry instrumental performance variation from day to day is a recognised problem. This means that samples measured on different days may not be directly comparable, likely due to slight variations in equipment operating conditions, laboratory ambient conditions and minor contamination left from other analysis. The aim of this study was to investigate the impact of this instrumental variation using a NOM standard mixture and ways to overcome it.

To understand the Orbitrap MS instrumental variation from day to day, a freshly prepared NOM standard mixture was analysed on several days. The data files were compiled and analysed using Compound Discoverer software. The molecular weights observed were assigned to molecular formula using the software. As part of the data processing various strategies were explored to deal with batch effects, including data-driven normalisation, removal of data with lower relative abundance and Systematic Error Removal Using Random Forest (SERRF).

For the NOM standard mixture, there were differences in the assigned molecular formulas as well as the relevant abundances. A total of 940 molecular formula were found for all the NOM mixture standard runs, with 357 found in more than one sample run. However, the compounds that were present in only one or two sample runs tended to have lower relative abundance, and hence removing compounds with lower relative abundance may reduce the influence of instrumental variation. In general, the greatest commonality across the sample runs was seen in the region where the H/C ratio was between 0.5-1.5 and the O/C ratio was <0.5, which corresponds to the condensed hydrocarbons and lignin-like compounds. The various correction strategies showed various levels of effectiveness.