Using Deep Learning to Extrapolate Protein Expression Measurements

Abstract

Mass spectrometry (MS)-based quantitative proteomics experiments typically assay a subset of up to 60% of the ≈20 000 human protein coding genes. Computational methods for imputing the missing values using RNA expression data usually allow only for imputations of proteins measured in at least some of the samples. In silico methods for comprehensively estimating abundances across all proteins are still missing. Here, a novel method is proposed using deep learning to extrapolate the observed protein expression values in label-free MS experiments to all proteins, leveraging gene functional annotations and RNA measurements as key predictive attributes. This method is tested on four datasets, including human cell lines and human and mouse tissues. This method predicts the protein expression values with average R 2 scores between 0.46 and 0.54, which is significantly better than predictions based on correlations using the RNA expression data alone. Moreover, it is demonstrated that the derived models can be "transferred" across experiments and species. For instance, the model derived from human tissues gave a R 2 = 0.51 when applied to mouse tissue data. It is concluded that protein abundances generated in label-free MS experiments can be computationally predicted using functional annotated attributes and can be used to highlight aberrant protein abundance values.

Experimental reproducibility limits the correlation between mRNA and protein abundances in tumour proteomic profiles

“…Since there are orders of magnitude more tumour samples with quantified transcriptomes than proteomes, a number of efforts have been made to use machine learning to predict protein abundances from mRNA abundances (Barzine et al, 2020; Fortelny et al, 2017; Li et al, 2019; Yang et al, 2020). Recently the NCI-CPTAC DREAM proteogenomics challenge engaged the community to predict protein abundances of tumour profiles using their corresponding genomic and transcriptomic information (Yang et al, 2020).…”

“…Recently there have been a number of attempts to predict protein abundances from transcriptomic data that have achieved modest success (Barzine et al, 2020; Fortelny et al, 2017; Li et al, 2019; Yang et al, 2020). We found here that proteins that are more reproducibly measured across experimental replicates are better predicted using machine learning.…”

“…Recently there have been a number of attempts to predict protein abundances from transcriptomic data that have achieved modest success(Barzine et al, 2020;Fortelny et al, 2017;Li et al, 2019;Yang et al, 2020…”

Predicting missing proteomics values using machine learning: Filling the gap using transcriptomics and other biological features

“…Since the protein expression level is the most important factor for biological interpretation, and considering the limited sensitivity and stochastic sampling of proteomics in addition to the very low correlation of the mRNA/protein expression level, we considered the possibility of obtaining predicted protein expression levels from the integration of as many possible features available from several OMICs data. Although we recognized that methods such as match-between-runs (MBR) [22] , DART-ID [23] , and IceR [24] have already been developed (and their limitations [25] ), including a deep learning approach to extrapolate proteomics values from transcriptomics values [26] , none utilized a complex multiomics strategy to approach in a novel manner the limitations of proteomics.…”

Computational Proteomics: Focus on Deep Learning

“…report a deep learning‐based method to predict protein abundance for all proteins. [ 8 ] This method uses a neural network to integrate mRNA expression data and functional annotation data to make predictions. It can be applied to proteins that are not experimentally quantified in any samples; therefore, it is different from the typical missing value imputation methods that are designed for the imputation of partially missing data.…”