Artificial intelligence applications in biomedicine face major challenges from data privacy requirements. To address this issue for clinically annotated tissue proteomic data, we developed a federated deep learning approach (ProCanFDL), training local models on simulated sites containing data from a pan-cancer cohort (n = 1,260) and 29 cohorts held behind private firewalls (n = 6,265), representing 19,930 replicate data-independent acquisition mass spectrometry runs. Local parameter updates were aggregated to build the global model, achieving a 43% performance gain on the hold-out test set (n = 625) in 14 cancer subtyping tasks compared with local models and matching centralized model performance. The approach's generalizability was demonstrated by retraining the global model with data from two external, data-independent acquisition mass spectrometry cohorts (n = 55) and eight acquired by tandem mass tag proteomics (n = 832). ProCanFDL presents a solution for internationally collaborative machine learning initiatives using proteomic data, for example, for discovering predictive biomarkers or treatment targets while maintaining data privacy. SIGNIFICANCE:A federated deep learning approach applied to human proteomic data, acquired using two distinct proteomic technologies from 40 tumor cohorts across eight countries, enabled accurate cancer histopathologic subtyping while preserving data privacy. This approach will enable the privacy-compliant development of large-scale proteomic artificial intelligence models, including foundation models, across institutions globally.

The proteome provides unique insights into disease biology beyond the genome and transcriptome. A lack of large proteomic datasets has restricted the identification of new cancer biomarkers. Here, proteomes of 949 cancer cell lines across 28 tissue types are analyzed by mass spectrometry. Deploying a workflow to quantify 8,498 proteins, these data capture evidence of cell-type and post-transcriptional modifications. Integrating multi-omics, drug response, and CRISPR-Cas9 gene essentiality screens with a deep learning-based pipeline reveals thousands of protein biomarkers of cancer vulnerabilities that are not significant at the transcript level. The power of the proteome to predict drug response is very similar to that of the transcriptome. Further, random downsampling to only 1,500 proteins has limited impact on predictive power, consistent with protein networks being highly connected and co-regulated. This pan-cancer proteomic map (ProCan-DepMapSanger) is a comprehensive resource available at https://cellmodelpassports.sanger.ac.uk.

Summary The proteome provides unique insights into biology and disease beyond the genome and transcriptome. Lack of large proteomic datasets has restricted identification of new cancer biomarkers. Here, proteomes of 949 cancer cell lines across 28 tissue types were analyzed by mass spectrometry. Deploying a clinically-relevant workflow to quantify 8,498 proteins, these data capture evidence of cell type and post-transcriptional modifications. Integrating multi-omics, drug response and CRISPR-Cas9 gene essentiality screens with a deep learning-based pipeline revealed thousands of protein-specific biomarkers of cancer vulnerabilities. Proteomic data had greater power to predict drug response than the equivalent portion of the transcriptome. Further, random downsampling to only 1,500 proteins had limited impact on predictive power, consistent with protein networks being highly connected and co-regulated. This pan-cancer proteomic map (ProCan-DepMapSanger), available at https://cellmodelpassports.sanger.ac.uk , is a comprehensive resource revealing principles of protein regulation with important implications for future clinical studies.

Abstract Motivation The output of electrospray ionization–liquid chromatography mass spectrometry (ESI-LC-MS) is influenced by multiple sources of noise and major contributors can be broadly categorized as baseline, random and chemical noise. Noise has a negative impact on the identification and quantification of peptides, which influences the reliability and reproducibility of MS-based proteomics data. Most attempts at denoising have been made on either spectra or chromatograms independently, thus, important 2D information is lost because the mass-to-charge ratio and retention time dimensions are not considered jointly. Results This article presents a novel technique for denoising raw ESI-LC-MS data via 2D undecimated wavelet transform, which is applied to proteomics data acquired by data-independent acquisition MS (DIA-MS). We demonstrate that denoising DIA-MS data results in the improvement of peptide identification and quantification in complex biological samples. Availability and implementation The software is available on Github (https://github.com/CMRI-ProCan/CRANE). The datasets were obtained from ProteomeXchange (Identifiers—PXD002952 and PXD008651). Preliminary data and intermediate files are available via ProteomeXchange (Identifiers—PXD020529 and PXD025103). Supplementary information Supplementary data are available at Bioinformatics online.

Reproducible research is the bedrock of experimental science. To enable the deployment of large-scale proteomics, we assess the reproducibility of mass spectrometry (MS) over time and across instruments and develop computational methods for improving quantitative accuracy. We perform 1560 data independent acquisition (DIA)-MS runs of eight samples containing known proportions of ovarian and prostate cancer tissue and yeast, or control HEK293T cells. Replicates are run on six mass spectrometers operating continuously with varying maintenance schedules over four months, interspersed with ~5000 other runs. We utilise negative controls and replicates to remove unwanted variation and enhance biological signal, outperforming existing methods. We also design a method for reducing missing values. Integrating these computational modules into a pipeline (ProNorM), we mitigate variation among instruments over time and accurately predict tissue proportions. We demonstrate how to improve the quantitative analysis of large-scale DIA-MS data, providing a pathway toward clinical proteomics.