Transcript
My name is Zhandong Liu. I'm an associate professor at the Baylor College of Medicine and also the chief data scientist at the Texas Children's Hospital. I was involved in a CZI project with Hugo Bellen and Shinya Yamamoto on how to speed and accelerate the diagnosis process.
My lab focused on developing AI algorithms. I was brought in this project to kind of figure out how to use AI to speed up the process.
Our algorithms takes the genome sequencing files from a patient along with their clinical files and process it using a tool called AI‑MARRVEL and will generate a small list of candidate variants and genes that can explain the phenotypes of the patient.
These algorithms were trained on a large amount of data that's been curated through Baylor Genetics, Baylor College of Medicine, The Undiagnosed Diseases Network. As you all know, that data is AI nowadays, right? When you are talking about high quality AI algorithms, oftentimes, those were trained on very good data sets.
We happen to have access to some of the top quality data sets that's curated at Baylor College of Medicine. These are about 4.5 million variants that have been reviewed by our board‑certified clinical geneticist. And based on those data sets along with the expert knowledge that we learned by interviewing those domain experts, we were able to create this state‑of‑the‑art artificial intelligent algorithm called AI‑MARRVEL.
We have used this algorithms on a couple projects. The first one is The Undiagnosed Diseases Network project where we were doing reanalysis. The second one is a project sponsored by CZI. It's called TMC‑CZI. A lot of the children that go through Texas Children's Hospital were sequenced and our algorithms was used to help them to do primary diagnosis and then secondary analysis as well.
The third project is the one that's sponsored by NIH called the Texome Project by Hugo Bellen and the Michael Wangler, where we were helping underserved population to gain access to whole exome sequencing, whole genome sequencing platforms and enable the AI diagnosis on this small population.
So, overall, the tool has been developed for almost two years and we have seen this tool being used in many groups and enabled to identify those very difficult variants used from hundreds or thousands of mutations.