OpenAI Enters the Stem Cell Game

The company has built an AI model tailor-made for stem cell development

According to the MIT Technology Review OpenAI has teamed up with Retro Biosciences, a biotech company designing cellular therapeutics, to create an AI model to assist stem cell researchers in the lab.

Background:

Sam Altman, founder of OpenAI, personally funded $180M Retro Biosciences in 2023. Retro is working on what appears to be induced pluripotent stem cells (IPCs).

What are IPCs? Essentially, you take a skin cell and coach it to become a stem cell. It was developed in Japan, and won a Nobel Prize in 2012.

Here's a short video that's easy to understand:

These IPCs can theoretically produce any tissue in the body, and don’t come with the political, ethical, or religious controversy like umbilical-derived cells do. Additionally, there’s a risk of your immune system rejecting allogeneic (sourced from someone else’s body) stem cells, since they came from autologous (sourced from your own body) sources.

What will the model do and why is it important?

The new model, dubbed GPT-4b micro, appears to help researchers accomplish this ‘coaching’ of cells by suggesting the next best steps based on previous research.

This technology has just started to hit other areas, one that I know a bit about, EV battery development. Making batteries requires a ton of materials science in a laboratory, often with tedious, repetitive, and difficult work.

I’ve interviewed several companies like Monolith AI, who developed a model to ‘suggest’ the next best step of battery development based on previous testing knowledge. Essentially, they load up this model with every battery chemistry textbook, research, and internal data available, making a superhuman researcher. The model then looks at what’s going on in the lab, and suggests what it will think will happen based on previous data, and suggest what to do next. Of course all of us known that AI hallucinates, so this isn’t a ‘set it and forget it’ sort of research robot, more of an adjunct tool to make some very lethal researchers.

In the EV battery realm, they’ve reduced testing time by upwards of 40%+ with these models, and some companies have claimed they’re applying AI models to condense years of research in just weeks. That’s huge.

I imagine this tech will have a similar setup.

What next?

Well, we don’t know. New R&D tech takes many years to hit the clinics if it ever does, and really this is just a new tool. It doesn’t mean every problem is magically solved, but I anticipate some big jumps in the next years as we begin to simulate bits and pieces of research using AI models.

Likely what will happen is they’ll make a handful of breakthroughs over the next year or two in a laboratory environment, and hopefully start clinical trials, which will take years, and then come up with some new medical breakthroughs.

We’ll see, but I’m excited to see it.