A team of cancer scientists led by Dr. Daniel De Carvalho, Principal Investigator of the Princess Margaret Cancer Center in Canada, combined with liquid biopsy, epigenetic changes and machine learning to develop a blood test that can detect and determine at the earliest stages of cancer and type of cancer.
The study, published in Nature on November 14th, not only describes a method for detecting cancer, but also brings a hope for early detection of cancer, because cancer is easier to treat and is far from the onset of symptoms at the earliest stages.
Dr. De Carvalho said: “At this stage, we are very excited about this. One of the main problems of cancer treatment is how to find it as soon as possible. This is a problem of fishing a needle in the sea because we have to find cancer specific variations which is only one billionth of DNA in the blood, especially at an early stage, when the amount of tumor DNA in the blood is very small.”
The research team identified tens of thousands of changes specific to each cancer through epigenetic changes rather than mutations. Then, through the big data approach, they use a classification tool developed by machine learning technology to identify the DNA from the tumor in the blood sample to determine the type of cancer. This turns the problem of “fishing needle in the sea” into a more easily solved problem of “fishing thousands of needles in the sea”. The computer only needs to find “several needles” to determine what cancer is.
The researchers compared 300 patient tumor samples from seven disease sites (lung cancer, pancreatic cancer, colorectal cancer, breast cancer, leukemia, bladder cancer, and kidney cancer) to samples from healthy donors by analyzing plasma circulatory cell-free DNA to track the origin and type of cancer. In each sample, “floating” plasma DNA matched the tumor DNA. The team later expanded the scope of the study. More than 700 tumor and blood samples from more cancer types have been analyzed and successfully matched.
In addition to being in the laboratory, the next step in validating this approach has been carried out in several countries, including the analysis of data from larger populations. These countries have collected blood samples from months to years before the diagnosis of cancer. This method will eventually be validated in a prospective study of cancer screening.
This method of combing clinical diagnosis with machine learning is inspiring.
Last modified: December 4, 2018