ML Approach Enables Early Cancer Detection

A new study has successfully combined genomics with machine learning to develop more accessible tests for earlier detection of cancer, researchers at the University of Wisconsin-Madison announced.

The technique, which uses readily available lab materials and analyzes fragments of cell-free DNA in blood plasma, detected cancers at an early stage among most of the samples studied.

The approach is based on the idea that DNA fragments from cancer cells might differ from healthy cell fragments in terms of where the DNA strands break and the nucleotides that surround these breaking points. The researchers used a technique called GALYFRE (Genome-wide Analysis of Fragment Ends) to analyze cell-free DNA from 521 samples, and sequenced data from an additional 2,147 samples from healthy individuals and patients with 11 different cancer types.

They developed a measure called the information-weighted fraction of aberrant fragments which they used, along with information on the DNA sequences surrounding fragment breaking points, to develop a machine-learning model that would compare DNA fragments from healthy cells to those from different types of cancer cells. The model accurately distinguished people with any stage of cancer from healthy individuals 91% of the time, and accurately identified samples from patients with stage 1 cancer in 87% of cases.

The researchers believe that GALYFRE holds promise for detecting cancer in early stages, and could also offer real-time efficacy assessment of ongoing treatment for certain types of cancer. However, more studies are needed to refine GALYFRE’s use in different age groups and in patients who have additional medical conditions. The team is also planning larger clinical studies to validate the test for specific cancer types such as pancreatic and breast cancer.