An integrated landscape of protein expression in human cancer

Abstract

Using public proteomics datasets, mostly available through the PRIDE database, we assembled a proteomics resource for 191 cancer cell lines and 246 clinical tumour samples, across 13 cancer lineages. We found that baseline protein abundance in cell lines was generally representative of tumours. However, when considering differences in protein expression between tumour subtypes, as exemplified in the breast lineage, many of these changes were no longer recapitulated in the cell line models. Integration of proteomics and transcriptomics data suggested that the level of transcriptional control in cell lines changed significantly depending on their lineage. Additionally, in agreement with previous studies, variation in mRNA levels was often a poor predictor of changes in protein abundance. To our knowledge, this work constitutes the first meta-analysis study including cancer-related proteomics datasets. We anticipate this aggregated dataset will be of significant aid to future studies requiring a reference to baseline protein expression in cancer.

Drug ranking using machine learning systematically predicts the efficacy of anti-cancer drugs

“…Therefore, DL methods may be better suited to train predictive models from phosphoproteomics and proteomics as larger data sets become available. Assessment of DRUML using external verification datasets from 53 cell lines analyzed by independent laboratories 44,45 revealed that around 85% of the drugs could be ranked with absolute errors < 0.15 and the drug rankings were statistically significant (by Spearman) within all cancer models tested (Figs. 5 and 6).…”

“…Drug sensitivity and RNA-Seq data were sourced from PharmacoDB 32 . Proteomics and phosphoproteomics data was generated in-house for 26 AML, 10 esophagus and 12 HCC cell lines in house (see above) or obtained from 44,45,25 . Drug response, proteomics and phosphoproteomics datasets were normalized and proteomics and phosphoproteomics data were further normalized by centering and scaling.…”

“…Colorectal phosphoproteomics validation data were obtained from PRIDE dataset identifier PXD001550 44 . Proteomics validation data were drawn from PRIDE dataset identifier PXD013455 45 . AML patient phosphoproteomics validation data was obtained from PRIDE project PXD005978 25 .…”

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

“…Previous work has found that certain pathways and processes are enriched in proteins that have higher or lower than average mRNA-protein correlation. For instance, ribosomal subunits have been found to have consistently lower than average mRNA-protein correlations across multiple studies (Clark et al, 2019; Mertins et al, 2016; Zhang et al, 2014, 2016) while members of pathways related to amino-acid metabolism have been found to have higher than average mRNA-protein correlation (Clark et al, 2019; Huang et al, 2021; Jarnuczak et al, 2021; Mertins et al, 2016; Zhang et al, 2014, 2016). This variation across functional groups has been attributed to differential post-transcriptional regulation.…”

An integrated view of baseline protein expression in human tissues

“…[4-6]) and the establishment of new data resources using re-analysed public datasets as the basis [7-9]. In this context of data reuse, the main interest of PRIDE is to disseminate and integrate proteomics data into popular added-value bioinformatics resources at the European Bioinformatics Institute (EMBL-EBI) such as Expression Atlas [10] (for quantitative proteomics expression data), Ensembl [10] (proteogenomics) and UniProt [11] (protein sequences information including post-translational modifications (PTMs)). The overall aim is to enable life scientists (including those who are non-experts in proteomics) to have access to proteomics-derived information.…”