Tempus AI and Daiichi Sankyo Bet on Multimodal AI to Sharpen ADC Development

Tempus AI is pushing deeper into drug development partnerships with a new strategic collaboration with Daiichi Sankyo, one centered on a problem that has become increasingly important in oncology: not just building better therapies, but identifying more precisely which patients are most likely to benefit from them. The agreement focuses on accelerating the clinical development and differentiation of an antibody-drug conjugate program by combining Daiichi Sankyo’s clinical trial and preclinical research data with Tempus’ real-world oncology data and AI models.

At the center of the collaboration is PRISM2, Tempus’ multimodal foundation model designed to analyze both pathology images and clinical data together rather than in isolation. That matters because oncology has become a data-dense field where images, molecular markers, patient histories, and treatment outcomes all carry pieces of the same puzzle. The premise here is that a model trained to interpret these inputs jointly may be able to generate more useful diagnostic and predictive insights than conventional approaches that treat each data stream separately. In practical terms, the two companies want to use AI to improve biomarker discovery, refine patient stratification, and make trial design more targeted from the outset.

The collaboration will begin with proof-of-concept AI models intended to optimize patient selection for a novel ADC and improve the probability of clinical success. That may sound like familiar pharma language, but it points to one of the central pressures in oncology R&D today: the need to match increasingly sophisticated therapies with narrower, more biologically appropriate patient populations. Rather than relying only on broad eligibility criteria, Tempus and Daiichi Sankyo aim to deploy these models across Tempus’ large oncology database to generate detailed response maps, which could help identify where the drug may show the strongest benefit and where future trials should focus. Those same tools may also help benchmark control arms more rigorously, which is not a small detail when regulators and investors alike are looking closely at trial design and comparative performance.

What makes this announcement notable is that it reflects a broader shift in how AI is being positioned in precision medicine. For a while, much of the conversation around AI in healthcare revolved around workflow efficiency, automation, or isolated prediction tasks. This partnership suggests a more ambitious use case: applying foundation models directly to the strategic architecture of clinical development. In that sense, AI is being cast less as a support layer and more as a way to shape how oncology programs are designed, differentiated, and de-risked before late-stage decisions become painfully expensive.

Ryan Fukushima, CEO of Data and Apps at Tempus, framed the deal as part of a larger evolution in drug development, arguing that multimodal AI and real-world data can help uncover unmet patient needs while identifying those most likely to respond to novel therapies. That is really the key idea behind the collaboration. ADCs are one of the hottest areas in oncology, but competition is intensifying and differentiation is getting harder. A better drug still needs a better development path, and increasingly, that path may depend on who gets selected, how trials are structured, and what signals can be surfaced earlier from messy real-world and clinical datasets.

For Daiichi Sankyo, the collaboration underscores how major biopharma companies are leaning into external AI platforms not just for discovery-stage experimentation but for clinically relevant decision support. For Tempus, it is another example of how its data assets and foundation models are being positioned as infrastructure for precision oncology. The bigger picture, maybe a little bluntly, is this: in modern cancer drug development, the contest is no longer only about making the molecule work. It is also about making the data work harder around it.