Have a hunch about cancer-driving variants hiding in non-coding regions of the genome and need a way to find and prioritize these mutations?

Researchers funded by NCI’s Division of Cancer Biology are testing a platform that could help bring those non-coding drivers into better focus.

Their platform, which blends statistical and deep learning models (including a generalized linear model and one using a neural network), can help you identify non-coding regulatory regions that drive cancer progression. Although trained on prostate cancer (in particular, metastatic castration-resistant) genomes, the authors say you could use the platform to look at other human cancers as well.

Corresponding author, Dr. Hosseinali Asgharian, of the University of California–San Francisco (UCSF), said, “Using our platform, we were able to combine whole-genome sequencing and matched RNA sequencing data to predict and prioritize functionally relevant mutations to better understand how these gene changes impacted gene expression control.”

Co-corresponding author, Dr. Felix Feng, also at UCSF, added, “Having a platform that prioritizes these mutations is key. By integrating computational and experimental tools (such as CRISPRi and MPRAs), our model paints a fuller picture of these cancer driving events.”

Senior author, UCSF’s Dr. Hani Goodarzi, noted, “Our study confirms that the platform works well in a targeted cohort with a specific cancer type. Moving forward, we’re planning to apply this integrated framework to find significant non-coding drivers in other cancer patient populations.”