MI Genomic Profiling Similarity Score Launched to Diagnose Cancer Unknown Primary Cases

David Spetzler, MS, MBA, PhD MI Genomic Profiling Similarity (GPS) Score, the latest addition to the comprehensive genomic profiling armamentarium at Caris Life Sciences, has been launched, according to a press release from the company.1 MI GPS Score, an artificial intelligence (AI)-driven tumor type biology similarity score, uses more than 6500 mathematical models to compare molecular characteristics in a tumor with an extensive database in order to provide new insights into the molecular subtype of cancer unknown primary (CUP), atypical, and other difficult-to-treat cases in cancer. These insights will help guide treatment decisions in these particular patient cases. A subset of results from Caris' Molecular Intelligence platform, a proprietary offering of the company, was used to develop the MI GPS Score. This tool will help to manage CUP or atypical cases, or those with clinical ambiguity. A team of pathologists reviewed the case with the additional information provided by the MI GPS Score, which compares molecular characteristics of the patient&rsquo;s tumor against that of the Caris database.For example, a tumor sample may have a similar molecular signature as select lung cancer cases found in the database, which would ultimately impact the diagnosis of this patient's tumor. According to a poster at the 2019 American Society of Clinical Oncology (ASCO) Annual Meeting,samplesof tumors that were classified with the MI GPS Score had over 95% accuracy. The vast majority of these tumors were CUP cases. According to the study investigators, in 5% to 10% of cancers, ambiguity is high enough that no tissue of origin can be determined, whereas the sample is labeled as a CUP. Without reliable classification of these tumors, physicians struggle to make treatment decisions in a timely manner. Approximately55,780 samplesfrom patient tumors with next-generation sequencing data were used to create a multiple parameter tumor type specific classification system, utilizing an advanced machine learning approach. The dataset was split 50% for training and 50% for testing for each classifier. Each classifier was trained to identify the cases that were similar to other cases of other histological origins.2 Overall, the tumor lineage classifiers were able to predict the correct classifiers in which the primary site was known, with accuracy ranging from 85% to 95%. In the group of patient cases that were identified as CUP (n = 500), the case could be predicted 100% of the time. The algorithms used in this test were able to aide in the diagnosis of unknown tumors. Investigators concluded the predictors rendered a histologic diagnosis to CUP cases. As a result, the test can help inform treatment decisions and ultimately improve outcomes in patients with CUP or atypical cases. &ldquo;Caris has the world&rsquo;s largest and most comprehensive database of paired molecular information with clinical outcomes. We are actively employing advanced machine learning capabilities with this data set to identify unique molecular signatures that our industry-leading pathologists can use to better identify cancer subtypes and predict patient response to certain therapies. The combination of AI and human intelligence provides the most comprehensive analysis available today to characterize a patient&rsquo;s tumor,&rdquo; said David Spetzler, MS, MBA, PhD, President and Chief Scientific Officer of Caris, in a statement. References: Caris Life Sciences Launches First Ever Molecular AI Product: A Clinical Genomic Profiling Similarity Score. Caris Life Sciences press release.https://www.carislifesciences.com/news/caris-life-sciences-launches-first-ever-molecular-ai-product-a-clinical-genomic-profiling-similarity-score/. Published December 11, 2019. Accessed December 16, 2019. Abraham J, Heimberger AB, Gatalica Z, et al. Machine learning algorithm analysis using a commercial 592-gene NGS panel to accurately predict tumor lineage for carcinoma of unknown primary (CUP). Posted presented at: 2019 ASCO Annual Meeting; May 31-June 4, 2019; Chicago, IL.