Pan-cancer machine learning predictors of primary site of origin and molecular subtype

Abstract

15Background: It is estimated by the American Cancer Society that approximately 5% of all 16 metastatic tumors have no defined primary site (tissue) of origin and are classified as cancers of 17 unknown primary (CUPs). The current standard of care for CUP patients depends on 18 immunohistochemistry (IHC) based approaches to identify the primary site. The addition of post-19 mortem evaluation to IHC based tests helps to reveal the identity of the primary site for only 20 25% of the CUPs, emphasizing the acute need for better methods of determination of the site of 21 origin. CUP patients are therefore given generic chemotherapeutic agents resulting in poor 22 prognosis. When the tissue of origin is known, patients can be given site specific therapy with 23 significant improvement in clinical outcome. Similarly, identifying the primary site of origin of 24 metastatic cancer is of great importance for designing treatment. 26Identification of the primary site of origin is an import first step but may not be sufficient 27 information for optimal treatment of the patient. Recent studies, primarily from The Cancer 28Genome Atlas (TCGA) project, and others, have revealed molecular subtypes in several cancer 29 types with distinct clinical outcome. The molecular subtype captures the fundamental 30 mechanisms driving the cancer and provides information that is essential for the optimal 31 treatment of a cancer. Thus, along with primary site of origin, molecular subtype of a tumor is 32 emerging as a criterion for personalized medicine and patient entry into clinical trials. 34However, there is no comprehensive toolset available for precise identification of tissue of origin 35 or molecular subtype for precision medicine and translational research. 37Methods and Findings: We posited that metastatic tumors will harbor the gene expression 38 profiles of the primary site of origin of the cancer. Therefore, we decided to learn the molecular 39 characteristics of the primary tumors using the large number of cancer genome profiles 40 available from the TCGA project. Our predictors were trained for 33 cancer types and for the 11 41 cancers where there are established molecular subtypes. We estimated the accuracy of several 42 machine learning models using cross-validation methods. The extensive testing using 43 independent test sets revealed that the predictors had a median sensitivity and specificity of 44 97.2% and 99.9% respectively without losing classification of any tumor. Subtype classifiers 45 achieved median sensitivity of 87.7% and specificity of 94.5% via cross validation and 46 presented median sensitivity of 79.6% and specificity of 94.6% in two external datasets of 1,999 47 total samples. Importantly, these external data shows that our classifiers can robustly predict the 48 primary site of origin from external microarray data, metastatic cancer data, and patient-derived 49 xenograft (PDX) data. 50 51 Conclusion: We have demonstrated the utility of gene expression profiles to solve the 52 important clinical challenge...

A Study of High-Grade Serous Ovarian Cancer Origins Implicates the SOX18 Transcription Factor in Tumor Development

“…''bulk'' molecular profiling (Flynn et al, 2018;Moran et al, 2016;Staub et al, 2010;Søndergaard et al, 2017).…”

Imprecise Data and Their Impact on Translational Research in Medicine

“…An effort in the direction of precision cancer therapy needs to address the knowledge improvement with regards to primary site of origin and accurate subtyping (in 2-5% of metastatic cancer patients cannot be located the primary site, thus left with a classified cancer of unknown primary and poor prognosis to only empiric treatment and insurgence of comorbidities). Flynn et al (26) built pan-cancer classifiers to predict multiple cancer primary site of origin from metastatic tumor samples.…”

RNA-Seq-Based Breast Cancer Subtypes Classification Using Machine Learning Approaches

“…More importantly, the decision rules discovered in this research can provide reference guidelines for diagnosis and drug development of lung cancer subtypes. Flynn et al [ 10 ] have studied several machine learning approaches, including KNN, random forest, and SVM, using gene expression data to determine the molecular subtypes of cancer. Hijazi and Chan [ 11 ] proposed a classification framework for cancer subtypes based on gene expression data.…”