The overall objective of the ProCan® cancer proteomics program is to develop proteomic methodologies and a knowledge base of clinical data and cancer tissue proteomes to enable the routine use of proteomics in the cancer clinic as a diagnostic and treatment decision-making aid [1]. We have developed a “near-universal” single-tube preparative technique that includes heating, disruption with zirconium beads, pressure cycling, trypsin/LysC digestion, chemical modification and solid phase extraction. This produces peptide extracts from cancer model systems and cancer tissues that have been stored fresh-frozen (FF), embedded in OCT, or as formalin-fixed paraffin-embedded (FFPE) tissue blocks. We currently use six SCIEX TripleTOF 6600 mass spectrometers operating continuously with >70% uptime and generate proteomic data by Data-Independent Acquisition (DIA) Mass Spectrometry (MS). The total processing time required to generate peptide and protein matrices is <9 hours from receipt of a tissue section, which will enable proteomic data to be delivered within a clinically-relevant time frame. The proteomic data are reproducible from instrument to instrument and over time [2], and resource availability is therefore the only limit to pipeline scalability. A pan-cancer database is being constructed through a large series of collaborative projects, each of which is designed to address a clinical problem, e.g., finding a multi-protein biomarker (“protein signature”) in pre-treatment cancer samples that predicts response to treatment. We typically use separate discovery and validation cancer sample cohorts from different treatment centres; examples will be presented. To date, approximately 17,000 cancer tissue proteomes have been generated. Projects involving cancer samples can usually analyse outcomes from only a single type of treatment or treatment combination, whereas cancer model systems can generate data for large numbers of treatments. We recently analysed a pan-cancer set of 949 cancer cell lines from Wellcome Sanger Institute that have been treated with >600 compounds and demonstrated the utility of the proteomic data for predicting drug response [3]. We have shown that machine learning models using these and other cancer cell line proteomes as training data can predict the cancer type of tumour samples; the accuracy of the predictions increased when tumour proteomes were added to the training set. This provides preliminary support for the notion that proteomic data from cell lines and other cancer model systems will have clinical utility. Additional steps required to enable the adoption of proteomics as a routine clinical tool will include prospective studies, health economic analyses, and obtaining regulatory approval.