Next-Generation Machine Learning Tool Detects Mutational Signature Linking Bladder Cancer to Tobacco Smoking

For the first time, a pattern of DNA mutations that connects bladder cancer to smoking tobacco has been discovered by scientists at the University of California San Diego (UCSD). This discovery was possible because of a robust new machine learning tool that the team designed to detect patterns of mutations produced by carcinogens and other DNA-changing processes.

The work, illustrated in the September 23^rd, 2022 issue of Cell Genomics, could help scientists classify what environmental aspects, such as exposure to UV radiation and tobacco smoke, cause cancer in some patients.

Each of these environmental exposures uniquely modifies DNA, producing a particular pattern of mutations known as a mutational signature. If a signature is detected in the DNA of the cancer cells of a patient, cancer can be mapped back to the exposure that formed the signature. Knowing which mutational signatures exist could also result in a more tailored treatment for the particular cancer afflicting a patient.

In this research, the team discovered a mutational signature in the DNA of bladder cancer that is connected to the smoking of tobacco. The finding is noteworthy as a mutational signature from tobacco smoking has been identified in lung cancer but never in bladder cancer.

There is strong epidemiological evidence tying bladder cancer to tobacco smoking. We even see a specific mutational signature in other tissues—such as the mouth, esophagus, and lungs—that are directly exposed to tobacco carcinogens. The fact that we weren’t finding this signature in the bladder was strange.

Ludmil Alexandrov, Study Senior Author and Professor, Bioengineering and Cellular and Molecular Medicine, University of California San Diego

Alexandrov and contemporaries currently demonstrate that there is a mutational signature from smoking tobacco in bladder cancer, and it is different from the signature detected in lung cancer. Furthermore, they demonstrate that this signature is also detected in regular bladder tissues of tobacco smokers with no sign of bladder cancer. The signature was not detected in the bladder tissues of people who do not smoke.

What this signature tells us is that certain mutations in your DNA are due to exposure to tobacco smoke. It doesn’t necessarily mean that you have cancer. But the more you smoke, the more mutations accumulate in your cells, and the more you increase your risk for developing cancer.

Marcos Diaz-Gay, Study Co-First Author and Postdoctoral Researcher, Alexandrov’s Lab, University of California San Diego

Made Possible by Next-Generation Machine Learning

The scientists discovered the tobacco signature using a next-generation machine learning tool designed by Alexandrov’s lab. The researchers state it is an extremely progressive, automated bioinformatics tool for deriving mutational signatures instantly from large quantities of genetic data.

“This is a powerful machine learning approach to recognize patterns of mutations and separate them from genomic data,” said Alexandrov. “It takes those patterns and deciphers them, so that we can see what the mutational signatures are and match them with their meaning.”

He equated the machine learning method to selecting particular individual discussions at a cocktail party.

“You have multiple groups of people talking all around you, and you are only interested in hearing certain individuals speaking,” Alexandrov said.

Our tool essentially helps you do that, but with cancer genetic data. You have multiple people around the world exposed to different environmental mutagens, and some of those exposures are leaving imprints on their genomes. This tool goes through all that data to pick out what are the processes that cause the mutations.

Ludmil Alexandrov, Study Senior Author and Professor, Bioengineering and Cellular and Molecular Medicine, University of California San Diego

The tool was employed to examine 23,827 sequenced human cancers. It detected four mutational signatures—including the one in bladder cancer linked to tobacco smoking—that had not been spotted by any other tool. The three other signatures, found in colon, stomach, and liver cancers, still require additional study to ascertain what processes triggered them.

To demonstrate the power of their tool, the scientists tested it against 13 prevalent bioinformatics tools. The tools were evaluated for their ability to derive mutational signatures from over 80,000 artificial cancer samples. The tool that Alexandrov’s team built had outclassed all the others. It spotted 20 to 50% more true positive signatures than the others, with five times fewer false positive signatures. It even did well when examining noisy data, whereas the other tools were unsuccessful.

“In bioinformatics, this is the first time that such a comprehensive benchmarking has been done on this scale for mutational signature extraction,” said Diaz-Gay. “It is a huge undertaking, comparing many tools across many datasets.”

Such an accomplishment is also costly, observed Alexandrov. “Thanks to funding from Cancer Research UK, we were able to do this technical, extensive evaluation, which isn’t commonly done.”

Creating a More User-Friendly and Personalized Tool

The researchers’ final goal is to develop a web-based tool that more scientists can utilize and consequently profile more patients.

Right now, this tool requires bioinformatics expertise to run it. What we want is to create a user-friendly version on the web, where researchers can just drop in a patient’s mutations, and it immediately gives you the set of mutational signatures and what processes caused them.

Ludmil Alexandrov, Study Senior Author and Professor, Bioengineering and Cellular and Molecular Medicine, University of California San Diego

“Our idea for the future is to leverage this tool to analyze patients on an individual level,” said Diaz-Gay.

This study received support from Cancer Research UK and the National Institutes of Health.

Journal Reference:

Islam, S. M. A, et al. (2022) Uncovering novel mutational signatures by de novo extraction with SigProfilerExtractor. Cell Genomics. doi.org/10.1016/j.xgen.2022.100179.