Cerebras And Mayo Clinic To Debut Genomic Tool

One of the most compelling use cases of AI is in healthcare, and over the past couple of years, the medical community has learned that there are a lot of different applications, not just one.

People often start by talking about improvements and diagnosis and treatment, and that is central to the clinical process. But AI can benefit patients in many other ways, too. There’s everything from billing and scheduling, to the critical use of real-time vital signs to look for risk.

However, focusing on those broader-spectrum goals, we have news from AI company Cerebras and the Mayo Clinic, announcing a joint project for AI tools used in patient care.

The technology was announced at the JP Morgan Healthcare Conference in San Francisco.

In this project, the stakeholders aim to use a human reference genome to combine with patient data in order to try to identify genetic differences against benchmarks, and apply that to treating patients.

The model was trained using data on 500 Mayo Clinic patients, and the genomic data.

A Quick Glossary

The Cerebras press release points out that system uses Mayo Clinic’s “patient exome data.” What is exome data?

Here’s what ChatGPT says:

“Exome data refers to information obtained by sequencing all of the protein-coding regions of the genome—known collectively as the ‘exome.’ These protein-coding regions (called ‘exons’) represent only about 1–2% of the entire human genome but contain the vast majority of known disease-related genetic variations. By focusing on the exome, researchers and clinicians can more efficiently identify disease-causing variants without having to sequence and analyze the entire genome.”

And what is the human reference genome?

Scientists describe it as a composite “idealized” version of a human genome, a sort of boilerplate template to serve as a control case for genomic variations.

What are foundational models?

Again, ChatGPT tells us:

“Foundational models are large-scale machine learning systems trained on massive datasets—often in a self-supervised manner—so they can learn general-purpose representations useful for a wide array of downstream tasks. They serve as a “foundation” because their broad learning enables them to be adapted or fine-tuned for specialized purposes (e.g., language translation, image classification, or question answering) without retraining from scratch. These models capture extensive patterns from text or other data forms, providing a versatile starting point for many applications, but they also bring challenges related to fairness, bias, and interpretability.”

Making Progress in Treating Rheumatoid Arthritis

Rheumatoid arthritis can be debilitating, and treatment results can be hard to measure.

Scientists involved in the Cerebras/Mayo Clinic project want to “shorten time to identify treatment and decrease the severity of the condition,” according to the press release, looking for “high performance against benchmarks.”

Here’s the relevant portion from the Mayo Clinic’s press release:

“Rheumatoid arthritis (RA) is a debilitating autoimmune disease, and the standard treatment approach often requires trials of different therapies to achieve disease remission. It can take several months to know if a therapy is working. A new genomic model developed by Mayo Clinic and Cerebras offers a potential solution to shorten the time to identify effective treatment and avoid long-term morbidity associated with the untreated disease. Early findings demonstrate high performance against benchmarks and show early promise in identifying patient response to therapy. As more patient data is added, the model's predictive power is expected to increase, leading to faster, more effective personalized treatment for RA patients.”

So the technology may lead to more improvements earlier on. Those familiar with arthritis know how crippling it can be, in a day to day way, with pain and other symptoms. So early intervention can be a powerful change in patient care for RA.

But that’s not the only kind of application for this technology, it could be used for conditions like cancer and heart disease, too.

Other Big AI Genomics Projects

There’s also a lot more going on in genomics.

There’s the NIH “All of Us” research program, where researchers want to collect dynamic data from over 1 million Americans, and use AI and machine learning to look for connections between genetics and disease.

There’s Deepmind’s research on protein structures, and work by medical companies to analyze molecular profiles for cancer research.

All of it contributes to the overall job of moving the ball forward on patient care with the latest AI tools and technologies available.

And then there are all of those wearables!

Having our vital signs at our fingertips will help us with self-care in so many important ways. Just think about how costly it currently is to order a patient a heart “holter monitor” and then think about what AI wearables will do!

Look for more on this soon.