Cancer proteomics: from identification of novel markers to creation of artifical learning models for tumor classification.

Studies of global protein expression in human tumors have led to the identification of various polypeptide markers, potentially useful as diagnostic tools. Many changes in gene expression recorded between benign and malignant human tumors are due to post-translational modifications, not detected by analyses of RNA. Proteome analyses have also yielded information about tumor heterogeneity and the degree of relatedness between primary tumors and their metastases. Results from our own studies have shown a similar pattern of changes in protein expression in different epithelial tumors, such as decreases in tropomyosin and cytokeratin expression and increases in proliferating cell nuclear antigen (PCNA) and heat shock protein expression. Such information has been used to create artificial learning models for tumor classification. The artificial learning approach has potential to improve tumor diagnosis and cancer treatment prediction.