Modern analytics in the life sciences has become very much dependent on (bio-)informatics to process and interpret the resulting data. But with the advent of machine learning models, this dependence has become a true partnership. Indeed, these machine learning models have proven so capable that they now singlehandedly succeed in improving the analytical capability of instruments, as is strikingly shown in mass spectrometry, in particular in mass spectrometry-based proteomics [1]. Here, models such as MS2PIP [2] and DeepLC [3], coupled to the MS2Rescore [4] variant of the Percolator rescoring engine, have been shown to boost identification performance in complex samples and/or protocols by upwards of 35% across the board, often performing even better than this baseline improvement.
This is nothing short of remarkable in itself, as these free software tools immediately boost the information recovery from both existing as well as newly acquired data sets, and are capable of this feat across instruments from different vendors and even different generations, performing at least as well on latest generation instruments as on older instruments.
Moreover, these approaches also allow completely new types of data analysis, providing novel insights into the underlying biology (and thus also pathology) encoded (but thus far unnoticed) in existing data sets. This additional exploratory power is of particular interest, as nearly all findings are new, and therefore of high fundamental as well as translational interest.
Furthermore, and this is probably the most exciting application of machine learning models in current biomedical applications, these models can reveal staggeringly detailed insight into pathologies at the molecular level [5], while also showing substantial promise for the mapping of entire networks of coordinated activity of (modified) proteins at a proteome-wide scale [6].
The resulting unique views on entire proteomes in their (dys)functional states will undoubtedly fuel novel developments, and possibly even some revolutions, in precision medicine, and should excite anyone in analytical life sciences about the possibilities offered by proteomics approaches today!
Literature:
[1] Neely et al., Journal of Proteome Research, 2023. [2] Declercq et al., Nucleic Acids Research, 2023. [3] Bouwmeester et al., Nature Methods, 2021. [4] Declercq et al., Molecular & Cellular Proteomics, 2022. [5] Claeys et al., Journal of Proteome Research 2023. [6] Koutrouli et al., Bioinformatics, 2024.
