Article Dans Une Revue Nature Communications Année : 2024

Imputation of label-free quantitative mass spectrometry-based proteomics data using self-supervised deep learning

Résumé

Imputation techniques provide means to replace missing measurements with a value and are used in almost all downstream analysis of mass spectrometry (MS) based proteomics data using label-free quantification (LFQ). Here we demonstrate how collaborative filtering, denoising autoencoders, and variational autoencoders can impute missing values in the context of LFQ at different levels. We applied our method, proteomics imputation modeling mass spectrometry (PIMMS), to an alcohol-related liver disease (ALD) cohort with blood plasma proteomics data available for 358 individuals. Removing 20 percent of the intensities we were able to recover 15 out of 17 significant abundant protein groups using PIMMS-VAE imputations. When analyzing the full dataset we identified 30 additional proteins (+13.2%) that were significantly differentially abundant across disease stages compared to no imputation and found that some of these were predictive of ALD progression in machine learning models. We, therefore, suggest the use of deep learning approaches for imputing missing values in MS-based proteomics on larger datasets and provide workflows for these.

Proteomics is a technology for the identification and quantification of proteins to answer a broad set of biological questions 1 and together with RNA and DNA sequencing offers a way to map the composition of biological systems. It is widely applied across many fields of research including identification of biomarkers and drug targets for diseases such as alcoholic liver disease (ALD) 2 , ovarian cancer 3 and Alzheimer's disease 4 . Different workflows have been developed for analysis of body fluids, cells, frozen tissues and tissue slides, and are rapidly evolving. Recent technological advancements have enabled proteome analysis at the single cell or single cell-population level 5,6 , allowing the selection of single cells using image recognition 7 . However, for most approaches, missing values are abundant due to the semi-stochastic nature of precursor selection for fragmentation and need to be replaced for at least some parts of the data analysis. Currently, imputation of missing values in proteomics data usually assumes that the protein abundance was below the instrument detection limit or the protein was absent. In general, the community differentiates between missing at random (MAR) which is assumed to affect all intensities across the dynamic range, whereas missing not at random (MNAR) becomes more prevalent the more the intensity of a peptide approaches the limit of detection of the instrument. However, not all missing values are due to this mechanism, and by assuming the limit of detection as the reason for missingness will lead to potentially wrong imputations and subsequently to biased statistical results that are limiting the conclusion from data. A strategy is therefore to combine missing completely at random (MCAR) and simulated MNAR in comparisons 8 .

Fichier principal
Vignette du fichier
WEBEL_2023.pdf (1.6 Mo) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte
Licence

Dates et versions

hal-04743757 , version 1 (18-10-2024)

Licence

Identifiants

Citer

Henry Webel, Lili Niu, Annelaura Bach Nielsen, Marie Locard-Paulet, Matthias Mann, et al.. Imputation of label-free quantitative mass spectrometry-based proteomics data using self-supervised deep learning. Nature Communications, 2024, 15 (1), pp.5405. ⟨10.1038/s41467-024-48711-5⟩. ⟨hal-04743757⟩
10 Consultations
5 Téléchargements

Altmetric

Partager

More