Background Colorectal cancer (CRC) is the third most common cancer worldwide and the fourth leading cause of cancer-related deaths. Genomic and transcriptomic classification system has been developed for heterogeneous CRC tumors, but their applications in clinical tissues are limited due to the degradability of mRNAs. No protein-based classification system has been reported.
Methods We analyzed the proteome of FFPE biopsy samples from 217 CRC patients with up to ~9 years survival using pressure cycling technology (PCT) and data-independent acquisition (DIA) mass spectrometry. Then we trained a model using deep neural network. An independent CRC cohort of 117 patients was further procured to validate the protein-based classifier.
Results We quantified > 8000 proteins from 490 FFPE samples including 88 biological replicates (r = 0.77) and 66 technical replicates (r = 0.95). Using machine learning technology, we established a novel and practical protein-based classification system, containing the expression of about 10 ten proteins, for CRC prognosis which was further verified in an independent validation cohort.
Conclusion We demonstrated the practicality of PCT-DIA for analyzing large number of FFPE biopsy samples from multiple cohorts and established a novel and practical protein-based classification system CRCprot for CRC prognosis.