Thank you @EACRnews for highlighting our recent publication in @NatureBiotech, where we used deep-learning to map somatic mutation rates and identify potential driver mutations genome-wide. Read it here

Delighted to have played a small role in this awesome paper revealing non-canonical open reading frames and microproteins in the human brain lead by @ErinDuffyLacy1 @briankalishMD and @MEGNeuro.

What happens when you merge AI, systems biology, and CRISPR? You find new ways to kill cancer! That's our mission at Serinus Bio. Checkout our Y Combinator launch page for more info DM me if you're interested in collaboration opportunities.

RT @CD_AACR: Read this week's Cancer Discovery #ResearchWatch, Genome-wide Mutation Rate Modeling Identifies Novel Driver Mutations, a summ….

📢Big news 📢: @AdamYaari and I are launching @SerinusBio, a therapeutics startup in @ycombinator S22. Our white-box AI platform supercharges the drug development pipeline by unraveling what a drug will do in the cells of real patients. DM me for collaboration opportunities!

RT @NatureBiotech: Genome-wide mapping of somatic mutation rates uncovers drivers of cancer

RT @EricTopol: Deep learning #AI to rapidly identify new driver mutations in 37 types of #cancer.by @MIT_CSAIL @la….

Finally a huge thank you to my co-authors @AdamYaari, @olivercpriebe, @FelixDietlein, Po-Ru Loh, and Bonnie Berger @lab_berger. This would not have been possible without them!.

More resources: 1) Everything you need to know about and use Dig 2) Browsable mutation rate maps 3) A great write-up from @MIT News

Why does this work matter? 1) It unleashes the power of deep-learning to understand the molecular biology of cancer. 2) It suggests ways to improve clinical sequencing to further benefit cancer patients. 3) It suggests possible therapeutic targets and strategies.

Finally, we looked for rarely-mutated genes that can still driver cancer. We found immense overlap between genes that are common drivers in one type of cancer but rare drivers in other types of cancer. Maybe oncogenes are slightly less tissue-specific than we thought!

We confirmed that mutations in the 5' UTR of the major tumor suppressor gene TP53 likely break that gene, thus driving tumor formation. We also found that mutations in the 5' UTR of the tumor suppressor ELF3 followed this same pattern, suggesting a new candidate noncoding driver!

There is strong evidence that cryptic splice mutations - those deep in the introns of genes that alter how a mRNA is spliced - are under positive selection in tumor suppressor genes. In fact, they likely account for ~5% of all single-base mutations that break tumor suppressors!

We then applied these maps to patient cohorts that were either whole-genome sequenced, whole-exome sequenced, or targeted-sequenced to find both coding and noncoding drivers of cancer. Here are some of our findings:.

We created somatic mutation rate maps for 37 different types of cancer. These maps can be interactively browsed thanks @pkerpedjiev and the power of @higlass_io. Take a look!

Accuracy, flexibility, and speed are foremost. Dig can capture 98% of variation in mutation rates at the 1Mb scale. Users can customize searches with mutation-level precision. Dig searches >10 million mutations across >100k genomic loci in <90 seconds on a personal computer.

So we developed Dig, a probabilistic deep-learning method that maps cancer-specific somatic mutation rates genome-wide. It then leverages these maps to search for evidence of cancer-causing mutations across arbitrary genomic locations.

With the rapid rise of whole-genome sequencing, we wanted to empower researchers and clinicians to find mutations contributing to cancer anywhere in the genome across any set of patients.

Somatic mutations acquired during life are responsible for cancer. Most efforts to find these underlying genetic drivers of cancer have focused on the 2% of the genome that codes for proteins. What about the other 98%?.

So excited to share our paper "Genome-wide mapping of somatic mutation rates uncovers drivers of cancer" is out on @NatureBiotech (and open access)! Here's a🧵on what we did and what we found!.