Machine Learning Helps Repair Genetic Damage

Using machine learning techniques, researchers from Massachusetts General Hospital and the National Cancer Research Centre, have discovered how to repair genetic damage to prevent DNA mutations.  

According to the research team, these findings could assist in creating new cancer therapies, advancing precision medicine.  

When there is DNA damage, the cell activates a mechanism called DNA damage response that causes proteins to rapidly bind damaged DNA to send alarm signals, repairing the damage. The goal of chemotherapy is to kill tumor cells by causing DNA lesions, which causes cancer cell collapse and death.  

“By knowing how DNA lesions occur and how they are repaired, we will learn more about how cancer develops and how we can fight it. Any new discovery in DNA repair will help develop better cancer therapies, whilst protecting our healthy cells,” a member of the Metabolism and Cell Signaling Group, Bárbara Martínez, said in a press release. 

The team developed a new methodology that, with the assistance of a machine learning method, has allowed for a precise and detailed analysis.  

“Until now, one limiting factor in tracking DNA repair kinetics was the inability to process and analyze the amount of data generated from images taken by the microscope,” Martínez said.  

Researchers have used high-throughput microscopy allowing for the acquisition of thousands of pictures of cells after genetic damage.  

“In the first phase, they introduced more than 300 different proteins into the cells and evaluated in a single experiment whether they interfered with DNA repair over time. This technique has led to the discovery of nine new proteins that are involved in DNA repair,” the press release stated.  

Hoping to expand their research, the authors visually monitored the 300 proteins after generating genetic damage by adopting a classic DNA micro-irradiation technique.  

“We saw that many proteins adhered to damaged DNA, and others did just the opposite: they moved away from the DNA lesions. The fact that they either bind to or remove themselves from damaged DNA, to allow the recruitment of repair proteins to the lesion, is a common feature of DNA repair proteins. Both phenomena are relevant,” Martínez said. 

One of the proteins discovered was PHF20. According to the authors, the protein moves away from lesions within seconds after damage to facilitate the recruitment of 53BP1. Cells without PHF20 cannot repair their DNA properly, making them more sensitive to irritation. 

According to researchers, machine learning technology offers new opportunities to study DNA repair and manipulate it.  

“An advantage is that both platforms are very versatile and can be used to discover new genes or chemical compounds that affect DNA repair. We have evaluated hundreds of proteins in minimal time by using techniques allowing direct visualization of DNA repair,” Martínez said.