CAFE MOCHA: An Integrated Platform for Discovering Clinically Relevant Molecular Changes in Cancer-An Example of Distant Metastasis- and Recurrence-Linked Classifiers in Head and Neck Squamous Cell Carcinoma

JCO Clin Cancer Inform. 2018 Dec:2:1-11. doi: 10.1200/CCI.17.00045.


Purpose: With large amounts of multidimensional molecular data on cancers generated and deposited into public repositories such as The Cancer Genome Atlas and International Cancer Genome Consortium, a cancer type agnostic and integrative platform will help to identify signatures with clinical relevance. We devised such a platform and showcase it by identifying a molecular signature for patients with metastatic and recurrent (MR) head and neck squamous cell carcinoma (HNSCC).

Methods: We devised a statistical framework accompanied by a graphical user interface-driven application, Clinical Association of Functionally Established MOlecular CHAnges ( CAFE MOCHA;, to discover molecular signatures linked to a specific clinical attribute in a cancer type. The platform integrates mutations and indels, gene expression, DNA methylation, and copy number variations to discover a classifier first and then to predict an incoming tumor for the same by pulling defined class variables into a single framework that incorporates a coordinate geometry-based algorithm called complete specificity margin-based clustering, which ensures maximum specificity. CAFE MOCHA classifies an incoming tumor sample using either its matched normal or a built-in database of normal tissues. The application is packed and deployed using the install4j multiplatform installer. We tested CAFE MOCHA in HNSCC tumors (n = 513) followed by validation in tumors from an independent cohort (n = 18) for discovering a signature linked to distant MR.

Results: CAFE MOCHA identified an integrated signature, MR44, associated with distant MR HNSCC, with 80% sensitivity and 100% specificity in the discovery stage and 100% sensitivity and 100% specificity in the validation stage.

Conclusion: CAFE MOCHA is a cancer type and clinical attribute agnostic statistical framework to discover integrated molecular signatures.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Biomarkers, Tumor / genetics*
  • Computer Graphics
  • DNA Copy Number Variations
  • DNA Methylation
  • Gene Expression Profiling
  • Head and Neck Neoplasms / classification*
  • Head and Neck Neoplasms / genetics
  • Head and Neck Neoplasms / pathology*
  • Humans
  • Lymphatic Metastasis
  • Models, Statistical*
  • Mutation
  • Neoplasm Recurrence, Local / genetics
  • Neoplasm Recurrence, Local / pathology*
  • Prognosis
  • Software
  • Squamous Cell Carcinoma of Head and Neck / genetics
  • Squamous Cell Carcinoma of Head and Neck / secondary*
  • Transcriptome*


  • Biomarkers, Tumor