New models help identify mutations that cause cancer | MIT News

Cancer cells can have thousands of mutations in their DNA. However, only a handful actually accelerate the progression of cancer. The rest is just a ride.

Distinguishing these harmful driver mutations from neutral passengers may help researchers identify better drug discovery targets. To support these efforts, the MIT-led team has built a new computer model that can quickly scan the entire genome of cancer cells and identify mutations that occur more frequently than expected. This suggests that it promotes tumor growth. This type of prediction was difficult because passenger mutations are so frequent in some genomic regions that they drown out the actual driver’s signal.

“We have created a probabilistic, deep-learning method that allows us to obtain a truly accurate model of the number of passenger mutations that should be ubiquitous in the genome,” said a MIT graduate student. Says Maxwell Sherman. “Then we can look for regions throughout the genome that have an accumulation of unexpected mutations, suggesting that they are driver mutations.”

In their new study, researchers found additional mutations throughout the genome that may contribute to tumor growth in 5 to 10 percent of cancer patients. Researchers say the findings could help doctors identify drugs that are likely to treat those patients well. Currently, at least 30% of cancer patients do not have a detectable driver mutation that can be used as a therapeutic guide.

Sherman, MIT graduate student Adam Yaari, and former MIT research assistant Oliver Prive, are the lead authors of the study and today Nature biotechnology.. Bonnie Berger, a professor of mathematics at MIT and head of the Computational Biology Group at the Institute for Computer Science and Artificial Intelligence (CSAIL), is a senior author of this study, along with Po-Ru Loh, an assistant professor at Harvard University. .. Associate member of Medical School and Broad Institute of MIT and Harvard. Felix Dietline, an associate professor at Harvard Medical School and Boston Children’s Hospital, is also the author of this paper.

New tool

Since the human genome was sequenced 20 years ago, researchers have been scrutinizing the genome to find mutations that contribute to cancer by causing cells to grow out of control and avoid the immune system. increase. This provided targets such as the epidermal growth factor receptor (EGFR), which is commonly mutated in lung tumors, and BRAF, a common driver for melanoma. Both of these mutations can be targeted by specific drugs.

Although these targets have proven useful, the genes that encode proteins make up only about 2 percent of the genome. The other 98% also contain mutations that can occur in cancer cells, but it is much more difficult to determine if any of these mutations contribute to the development of cancer.

“There was really a lack of computational tools that could search for these driver mutations outside the protein coding region,” says Berger. “That’s what we were trying to do here. We design the calculation method so that we can see not only 2% of the genome encoding the protein, but also 100%.”

To that end, researchers trained a kind of computational model known as deep neural networks to search the cancer genome for mutations that occur more frequently than expected. As a first step, they trained the model with genomic data from 37 different types of cancer. This allowed the model to determine the background mutagenesis of each of those types.

“The really cool thing about our model is that once you train for a particular type of cancer, you learn the mutation rates for that particular type of cancer at the same time, anywhere in the genome,” Sherman said. Mr. says. “You can then query the mutations found in the patient cohort against the expected number of mutations.”

The data used to train the model comes from the Roadmap Epigenomics Project and an international data collection called Whole Genome Pancancer Analysis (PCAWG). By analyzing the model of this data, researchers obtained a map of expected passenger mutation rates throughout the genome. This gives the genome the expected rate at any set of regions (up to a single base pair).

Change the landscape

Using this model, the MIT team was able to add to the known situation of mutations that could cause cancer. Currently, known drivers will appear in about two-thirds of the time when tumors in cancer patients are screened for mutations that cause cancer. New results from the MIT study provide possible driver mutations in an additional 5-10 percent of the patient pool.

One type of non-coding mutation that researchers have focused on is called a “potential splice mutation.” Most genes are composed of exon sequences that encode protein-building instructions and introns, which are spacer elements that are normally trimmed from messenger RNA before being translated into protein. Mysterious splice mutations are found in introns and can disrupt the cellular mechanisms that splice them. This will include introns when they shouldn’t.

Using their model, researchers found that many mysterious splice mutations appear to disrupt tumor suppressor genes. In the presence of these mutations, the tumor suppressor is accidentally spliced ​​and ceases to function, causing the cell to lose one of its defenses against cancer. The number of mysterious splice sites found by researchers in this study accounts for about 5 percent of the driver mutations found in tumor suppressor genes.

Targeting these mutations may offer new ways to potentially treat those patients, researchers say. One possible approach, still under development, is to patch mutations in DNA with the correct sequence using short strands of RNA called antisense oligonucleotides (ASOs).

“If we can eliminate the mutation in a way, we can solve the problem. These tumor-suppressing genes will continue to function and may possibly fight cancer,” says Yaari. “ASO technology is actively being developed and this can be a very good application.”

Another region where researchers have discovered high levels of non-coding driver mutations lies in the untranslated regions of some tumor suppressor genes. It has already been known that the tumor suppressor gene TP53, which is defective in many types of cancer, accumulates many deletions in these sequences known as the 5’untranslated regions. The MIT team discovered the same pattern with a tumor suppressor called ELF3.

Researchers also used their model to investigate whether commonly known mutations could cause different types of cancer. As an example, researchers have found that BRAF, previously associated with melanoma, also contributes to cancer progression at a lower rate than other types of cancer, such as the pancreas, liver, and gastroesophagus. ..

“That is, there is actually a lot of overlap between the general driver situation and the rare driver situation, which provides an opportunity for diversion of treatment,” says Sherman. “These results are guidelines for clinical trials that need to be set to allow these drugs not only to be approved for one cancer, but also for many cancers and to support more patients. May be. “

This study was partly funded by the National Institutes of Health and the National Cancer Institute.