Virtual patients move AI beyond early discovery—toward the clinical decisions where drug-development value is won or lost
BEIJING,China — DeepoMe and The Future Laboratory at Tsinghua University todayannounced the release of a joint preprint, “Toward a Self-Learning AI Agentfor Drug Repurposing: Building Human-Scale Representations for VirtualPatients.” The work explores a critically under-modeled frontier in AI forscience: human-scale modeling—the representations needed to reason abouthow human biological states change under intervention.
AI is rapidlytransforming how researchers model proteins, generate molecules and simulatecellular perturbations. But the central opportunity is no longer only to modelmolecules or cells; it is to model human response to intervention.
A recentNature Reviews Drug Discovery Perspective makes the economic case: reducingfailure by 20% at Phase II—the first major test of whether a medicine deliversmeaningful benefit in people—can save nearly $900 million per successfuldrug launch, while the same improvement at early candidate selection has acomparatively small effect on total development cost. The article also findsthat clinical programs using biomarkers to identify likely responders canreduce the average cost per successful launch to roughly half that ofunstratified programs. The next frontier is therefore human-scale modeling:representing which biological states matter, which patients may respond andwhat evidence should guide the next intervention decision.
DeepoMe isbuilding Human Response Intelligence: infrastructure for connectingmeasured human biological state, structured interventions and longitudinallyobserved response. The company is developing this work first in longevityscience and aging-related disease—settings that combine repeated measurement,multidimensional biological state, modifiable interventions and cross-diseaserelevance.
Human ResponseIntelligence starts with defining state: determining which biologicaldimensions best describe a person for a particular decision. DeepoMe organizesthese dimensions in a human-scale biological map. Modules arereusable, evidence-linked units of that map—such as aging processes, immunestate, organ function, nutrition and exposure history. A representationis the decision-ready combination of modules selected for a particular disease,intervention and response question.
SteeraMedBench is the evaluation and optimization engine for this map. It testswhich modules—and which combinations of them—improve intervention reasoning,then selects and integrates the combinations that add the strongest distinctvalue into optimized maps for the decision at hand.
In thereported analyses, SteeraMed Bench integrates 332 modules spanning aging, organsystems, immunity, nutrition and food-derived interventions. It evaluates theirutility across drug-repurposing tasks involving 1,916 small molecules andmultiple disease contexts. The results indicate that no single biological mapis universally optimal: different diseases require different combinations ofbiological coordinates. In other words, representation selection is itself ameasurable driver of decision quality.
The frameworkalso includes an LLM-assisted workflow for proposing and evaluating candidatemodules against explicit utility and redundancy criteria. For example, theHallmarks of Aging framework has expanded from nine hallmarks to twelve and nowfourteen major hallmarks. Such scientific frameworks have traditionally evolvedthrough expert synthesis of a growing literature. SteeraMed Bench adds aquantitative layer: it tests proposed hallmarks against explicit utility,coverage and redundancy criteria, helping scientists build stronger knowledgeframeworks by distinguishing dimensions that improve decision quality fromthose that add little beyond existing representations.
Thisinfrastructure provides the coordinate system for steerable biological worldmodels. It enables models to reason over explicit, testable representationsof human state, mechanism and intervention; combine general-purpose maps withtask-specific maps; and improve those representations as new evidence accumulates.
“AI forScience has become remarkably good at seeing molecules and cells,” said Dr.Jianghui Xiong, Chief Scientist of DeepoMe. “But before we can ask how anintervention changes a person, we must define the biological state that mattersfor that person and that decision. Human Response Intelligence has two tasks:first, identify the dimensions that best represent human state; second, learnhow that state may change under intervention. SteeraMed Bench makes the firsttask measurable by evaluating and optimizing modules into the maps that providethe strongest basis for intervention reasoning.”
For DeepoMe,longevity science and aging-related disease are the initial learning domain forthis AI mission. These settings make it possible to observe human biologyrepeatedly, characterize state across multiple dimensions, document modifiableinterventions and study response over time.
DeepoMe’smultidimensional aging-assessment work, including Capome, is an entrypoint for measuring human biological state longitudinally and interpretably. Bylinking repeated measurements with structured intervention exposure andfollow-up, it creates the conditions to learn how—and for whom—an interventionchanges that state.
For researchand biopharma partners, DeepoMe is developing an auditabletranslational-decision workflow for prioritizing repurposing hypotheses,defining candidate responder populations and specifying the next validationexperiment or longitudinal study.
For aging andlongevity science, this infrastructure can move intervention reasoning frompopulation-level associations toward individualized N-of-1 predictionand, ultimately, the generation of individualized intervention programs. Itidentifies the biological dimensions most relevant to a given person, prioritizesintervention hypotheses and pairs them with longitudinal follow-up to learnwhether that person’s state changes as predicted.
The preprintis available at https://www.preprints.org/manuscript/202608.0998(DOI: 10.20944/preprints202608.0998.v1). SteeraMed Bench can be explored at https://steeramed.com/bench.
About SteeraMed Bench
SteeraMedBench is a research framework and evaluation-and-optimization engine forhuman-scale biological maps. It tests reusable, evidence-linked modules andtheir combinations to identify the representations that add the strongestdistinct value for a particular intervention decision. By measuring utility,coverage, redundancy and incremental value in drug-repurposing andintervention-reasoning tasks, it produces optimized maps for hypothesisgeneration, validation planning and future Human Response Intelligence systems.
About DeepoMe
DeepoMe is aBeijing-based research and technology company building Human ResponseIntelligence: infrastructure for connecting measured human biological state,structured interventions and longitudinally observed response. The companydevelops these capabilities first in longevity science and aging-relateddisease, through work spanning multidimensional state measurement,interpretable biomedical representations, intervention reasoning andlongitudinal learning. Its aim is to improve the quality, auditability andtestability of biomedical and translational hypotheses.
Media Contact
DeepoMe Media Relations
Email: [email protected]
Website: https://deepome.com