AI-based predictive biology coming soon, Foresite’s Bajaj says
CEO of Foresite Labs betting on integrating AI/MI with medical product R&D — The BioCentury Show
Clinical trials will be launched in the next year or two to test predictions about causal biology made by artificial intelligence models, Vik Bajaj, co-founder and CEO of Foresite Labs and managing director of Foresite Capital Management, believes.
Foresite teamed up with Arch Venture Partners last year on a $1 billion venture round to fund Xaira Therapeutics Inc., which aims to reinvent the R&D process via AI-driven protein design and biological discovery technologies.
Bajaj, who serves as interim president of Xaira, discussed the potential for artificial intelligence to accelerate biomedical progress in an interview with The BioCentury Show.
He noted AI and machine learning are already being used to design therapeutic agents, and said the next frontier is using the technology to create insights into causal biology that can be used to create investigational drugs.
The challenge is on the biology side of the equation — generating and collecting data — and not on the technology side.
“If you take all such data that the world has produced, as promising as it is, and put it together, it’s not enough to build models of biology yet.”
“Data is definitely a competitive moat,” Bajaj said. Meanwhile, as the unveiling of DeepSeek demonstrates, AI models are becoming less expensive and less of a competitive advantage.
But while all the information needed to create a large language model company at scale is available in the public domain, there isn’t enough data available right now to create predictive AI-based models of human biology, he noted.
“If you take all such data that the world has produced, as promising as it is, and put it together, it’s not enough to build models of biology yet,” Bajaj said, that can “teach us things that we can’t learn from existing human datasets.”
He added, however, that Xaira and other companies are on the verge generating datasets that are massive enough to test whether it is possible to “train and build a machine learning model that has some predictive value in telling you the association, and ideally the causal link, between a set of targets and a human disease.”
While there is a great deal of interest in applying AI and machine learning to the biopharma space, there are competing theories about which business models will be successful. One of the divides is between companies that are focused on creating tools that biopharma companies can use, and those that are developing those tools and integrating them into discovery and development programs.
Bajaj is in the later camp.
“It is hard to create a machine learning company” for the life sciences “without the crucible of product development in the company,” he told BioCentury. “If you do that, you begin to ask the wrong questions and spend a lot of time answering things that may not in the end accelerate the pace with which effective therapies get to patients.”
Foresite and Xaira are betting that machine learning companies that develop their own therapeutics pipelines “will win, even in the core R&D, and that there is a virtuous cycle between all of those activities — that not only can they co-exist, but it’s healthy for them to co-exist in the same company,” he said.