Novel Machine Learning Blood Tests Detects Cancers

Researchers at the Johns Hopkins Kimmel Cancer Center are developing a novel blood testing tool that combines genome-wide sequencing of single molecules of DNA lost from cancers and machine learning to allow for earlier identification of lung and other malignancies.

The GEMINI (Genome-wide Mutational Incidence for Non-Invasive Cancer Detection) test looks for alterations in DNA throughout the genome. First, a blood sample is obtained from a cancer-risk individual. The plasma is then collected and cell-free DNA (cfDNA) released by malignancies is sequenced utilizing cost-effective whole genome sequencing. Single molecules of DNA are examined for sequence changes and utilized to generate mutation profiles across the genome. Finally, to distinguish persons with cancer from those who do not, a machine learning model trained to recognize variations in cancer and non-cancer mutation frequencies in distinct regions of the genome is used. The classifier provides a value ranging from 0 to 1, with a higher score indicating a greater likelihood of malignancy.

In a series of GEMINI laboratory tests, researchers discovered that the method, when combined with computed tomography imaging, diagnosed over 90% of lung tumors, including those in patients with stage I and II disease. The experiment, described as a proof-of-concept study, will be published online on July 27 in the journal Nature Genetics.

“This study shows for the first time that a test like GEMINI, incorporating genome-wide mutation profiles from single molecules of cfDNA, in combination with other cancer detection approaches, may be used for early detection of cancers, as well as for monitoring patients during therapy,” says senior study author Victor Velculescu, M.D., Ph.D., professor of oncology and co-director of the cancer genetics and epigenetics program at the Kimmel Cancer Center.

The study mostly focused on detection of lung cancer in high-risk populations, says Daniel Bruhm, lead study author and graduate student in the human genetics program at the Johns Hopkins University School of Medicine. “However, we detected altered mutational profiles in cfDNA from patients with other cancers, including liver cancer, melanoma or lymphoma, suggesting it may be used more broadly,” Bruhm says.

GEMINI was developed after researchers analysed whole-genome sequences of tumors from 2,511 persons from the Pan-Cancer Analysis of Whole Genomes project, identifying differential mutation frequencies across the genome in different tumor types. Lung tumors, for example, were found to have an average of 52,209 somatic mutations per genome. The researchers also identified genomic regions with the highest number of mutations, discovering that high-frequency mutational sites in tumor tissue and blood-derived cfDNA from individuals with lung cancer, melanoma, or B cell non-Hodgkin lymphoma were identical.

GEMINI’s capacity to detect sequence variations in cfDNA from 365 participants in a prospective observational cohort (LUCAS), including people at high risk of lung cancer, was tested. GEMINI scores were greater in cancer patients than in non-cancer patients. GEMINI was also paired with DELFI (DNA evaluation of fragments for early interception), a previously developed test that detects changes in the size and distribution of cfDNA fragments across the genome, to increase diagnosis of early-stage lung cancer. The combined method found some cancer samples that GEMINI missed. GEMINI coupled with DELFI accurately detected lung cancer in 91% of 89 samples from patients in the LUCAS cohort. A further validation cohort of 57 persons, the majority of whom had early-stage lung cancer, yielded similar results.

The researchers also looked at how GEMINI was used in other study samples, including seven individuals who had no identifiable malignancies at the time of blood collection. They had a GEMINI score of 0.78 on average, which was greater than persons who did not have cancer. Six people tested positive for lung cancer using GEMINI and were diagnosed between 231 and 1,868 days later, indicating that abnormalities in cfDNA mutation patterns can be detected years before traditional diagnoses.

Additional tests revealed that GEMINI could distinguish between lung cancer subtypes and detect early liver tumors. GEMINI scores reduced after the early response to medication in a group of patients getting lung cancer drug treatment, indicating that the testing could be used to follow patients during therapy.

Together, the results indicate that the combination of genome-wide GEMINI mutation analyses and DELFI fragmentation analyses of cfDNA “may provide an opportunity for cost-efficient, scalable detection of cancers,” says Rob Scharpf, Ph.D., associate professor of oncology at the Kimmel Cancer Center. Larger clinical trials are needed to validate the tool before it could become available for clinical use, he says.

Source Link