Advertisement

Machine Learning Algorithm for Genome-Derived Prediction of Tumor Type


Advertisement
Get Permission

In a study reported in JAMA Oncology, Penson et al developed a machine learning algorithmic classifier that may be successful in identifying tumor type and origin based on DNA-sequence data obtained at point of care.

The study used machine learning to construct and train an algorithmic classifier in a cohort of prospectively sequenced tumors of 22 cancer types from 7,791 patients with advanced disease.

“These results suggest that the application of artificial intelligence to predict tissue of origin in oncologic practice can act as a useful complement to conventional histologic review to provide integrated pathologic diagnoses, often with important therapeutic implications.”
— Penson et al

Tweet this quote

Key Findings

The correct tumor type was predicted for 5,748 (73.8%) of 7,791 patients in the training set and in 8,623 (74.1%) of 11,644 patients in an independent cohort. Overall, predictions in 3,388 cases (43.5%) were classified as high-confidence predictions, having a > 95% probability of being accurate. Informative molecular features and feature categories were found to vary widely by tumor type.

Analysis of circulating tumor DNA resulted in correct prediction of tumor type in 12 (63.2%) of 19 patients with genitourinary cancers, 23 (85.2%) of 27 patients with metastatic breast cancer, and 10 (71.4%) of 14 with metastatic prostate cancer. Probable tissue of origin was predicted from targeted tumor sequencing in 95 (67.4%) of 141 patients with cancers of unknown primary site.

Use of the classifier in patients in active care resulted in genome-directed change of diagnosis in two patients initially diagnosed with metastatic breast cancer. This led to a change in the selected therapy and achievement of clinical response.

The investigators concluded, “These results suggest that the application of artificial intelligence to predict tissue of origin in oncologic practice can act as a useful complement to conventional histologic review to provide integrated pathologic diagnoses, often with important therapeutic implications.”

Barry S. Taylor, PhD, and Michael F. Berger, PhD, of Memorial Sloan Kettering Cancer Center, are the corresponding authors for the JAMA Oncology article.

Disclosure: The study was funded by Illumina, grants from the National Cancer Institute, and others. For full disclosures of the study authors, visit jamanetwork.com.

The content in this post has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the ideas and opinions of ASCO®.
Advertisement

Advertisement




Advertisement