How Do You Train Genomics AI? On Natural Selection Itself, Says VP of Illumina’s AI Lab, Kyle Fahr

Chapters:

0:00 First-of-kind studies to utilize AI to decode variation in humans

3:00 Training genomic AI on natural selection

9:10 What sets Illumina’s algorithms apart?

11:40 Insights of PrimateAI-3D and SpliceAI

21:40 How will AI impact genomics long term?

In today’s show, we talk with Kyle Farh, the VP and Distinguished Scientist leading Illumina’s Artificial Intelligence Lab.

"We lack a way of training genomics AI as readily as you would for natural language processing, says Farh.  "There are vast amounts of data, but there are no labels, no supervision.  One of the most powerful tools that we’ve found is using clues from natural selection.

Farh says the company began the AI lab seven years ago when deep learning took off. Two of the team’s AI algorithms, PrimateAI-3D and SpliceAI, published in Cell and Nature Genetics, were among the first examples of deep learning algorithms in genomics to become widely used by clinical labs and researchers in interpreting variants of unknown significance.  

We take a close look at Illumina’s work in AI and delve into the long-term impact of AI on genomics.