Insilico Medicine’s VAC platform makes biological age a core variable as it attempts to model cellular change across six biological scales.
Insilico Medicine has previewed a Virtual Aging Cell platform designed to model not simply the state of a cell, but how that state changes with biological age – bringing temporal dynamics into computational models intended for target discovery, cellular reprogramming and longevity research.
The multi-agent platform, known as VAC, coordinates AI agents across six biological scales, from molecular and intracellular processes through intercellular signaling, tissues and organs to the organism level. Its distinguishing premise is that biological age should be a core condition within the model, allowing simulated cells and their surrounding systems to behave differently depending on where they sit along an aging trajectory.
Longevity.Technology: Biology has always had a rather awkward relationship with the snapshot. We sample a cell, sequence it, characterize its state and then – understandably enough – build models around what we have captured; but a living cell is not a still life. It is dividing, differentiating, responding to its neighbors, accumulating damage and changing its relationship with the tissue around it, while biological age alters the rules as it goes. That is what makes Insilico’s Virtual Aging Cell interesting: not simply that it brings agentic AI to cellular modeling, but that it makes biological age a first-class condition rather than an annotation somewhere near the edge of the dataset. VAC is an attempt to turn increasingly sophisticated photographs into something closer to a film.
For geroscience, that distinction is rather more than computational housekeeping. Aging is, almost by definition, a trajectory – molecular damage accumulates, cell identity shifts, tissue environments change and processes occurring at one level ripple into others – so a model that cannot accommodate time may describe an aged cell beautifully while telling us rather less about how it got there, or what might persuade it to take another route. Potentially more interesting still is the attempt to simulate how interventions might redirect cellular fate. But a convincing simulation is not the same thing as predictive biology; the real test will come when virtual trajectories survive prospective wet-lab validation and, ideally, tell researchers something they did not already know. AI is very good at learning from what biology has done; biology remains disconcertingly good at doing something else.
Giving the model a clock
Virtual cells are, in essence, biology’s answer to a flight simulator: AI and mathematical modeling standing in for the real thing, useful for predicting drug responses or hunting therapeutic targets without touching a Petri dish. The trouble is that most of them are still learning from a single frozen moment. A cell caught mid-differentiation, mid-reprogramming or mid-aging doesn’t behave like a cell that has arrived somewhere and stopped; it is still answering to its neighbours, its tissue, and whatever organism-level context happens to be running in the background. Model the snapshot and you get a very good photograph. Model the process and you get something considerably harder – and considerably more useful.
“Every computational model needs a first principle, and ours is biological age,” said Alex Zhavoronkov, founder, co-CEO and Chief Business Officer of Insilico Medicine.
“This has been a decade-long scientific journey, from raising an industry-defining question at NVIDIA GTC in 2014, to teaching AI the language of aging biology with the PreciousGPT series, and now reimagining the virtual cell through agentic AI swarms.”
That lineage matters. Precious1GPT combined DNA methylation and transcriptomic data to predict biological age; Precious2GPT moved into age-conditioned generation of synthetic multi-omics data, while Precious3GPT incorporated transcriptomic, methylation and proteomic information across humans, mice, rats and monkeys. VAC is the attempt to take those increasingly sophisticated representations of biological age and put them in motion. They also give the newer platform something to check its work against: when VAC’s agents need to judge whether a molecular signal genuinely looks younger, the PreciousGPT models supply the peer-reviewed reference profiles for comparison, rather than leaving that judgment to heuristic guesswork.
Agents at every level
Rather than compressing biological information into one model, VAC deploys specialist agents across molecular, intracellular, intercellular, tissue, organ and organism levels. Master Agents coordinate information in both directions, allowing molecular changes to propagate upward while organism-level context influences reasoning further down the hierarchy.
“Biology unfolds across time and interconnected levels, yet many computational models examine only one layer or moment in isolation,” said Petrina Kamya, Vice President, Global Head of AI Platforms and President of Insilico Medicine Canada.
“VAC is designed to bring those dimensions together, combining biological age with multi-agent reasoning to explore how changes in a cell’s age affect tissues, organs, and the broader organism. Our goal is to create a more dynamic environment for studying disease, evaluating interventions, and finding new opportunities for drug discovery.”
When younger is not rejuvenated
That architecture faces a particularly interesting test when VAC is asked to distinguish genuine rejuvenation from a transient molecular phenotype that happens to look younger – a familiar problem in a field where movement in an aging biomarker does not necessarily mean the underlying biological system has become functionally younger. That test spans all six of VAC’s scales, from molecular and intracellular processes through intercellular signaling, tissues and organs to the organism level. Each of those judgments is anchored against the PreciousGPT reference profiles established earlier – the system isn’t simply deciding a signal looks younger, it’s checking that signal against what younger molecular data has already been shown to look like.
Speaking to Longevity.Technology, the Virtual Aging Cell project team explained that age is treated as “a first-class input, not a derived output”, sitting within a shared organism context alongside sex and health status.
“When an intervention is simulated, the platform checks whether this context variable has stably shifted – not just whether expression profiles look younger at one time point,” the team said.
That test is then carried across VAC’s hierarchy. More than 30 Specialist Agents independently monitor its six biological scales; for a simulated intervention such as partial reprogramming to qualify as a genuine reset, the change must propagate coherently across all six, from the molecular level up to the organism.
“A genuine reset shows coherent changes in both directions – molecular agents detect epigenetic rejuvenation, organ/organism agents confirm corresponding functional shifts, and the context AGE updates,” the team told us.
A temporary fluctuation should fail that test. Molecular signals might acquire a younger profile, but if higher-level agents find no corresponding functional rejuvenation, biological age does not stably update; the Top-Down Master Agent then pushes the system back toward its baseline through negative feedback, rather than simply flagging the mismatch and moving on. Insilico says the check is unusual enough to be genuinely rare: a survey of 48 papers in the space, the company told us, turned up no existing system combining all the elements needed to run it.
From prediction to perturbation
Insilico ultimately wants VAC to simulate interventions including drug-target inhibition, gene knockout and environmental stimuli, allowing agents to reason through downstream pathway changes and cross-scale effects. The company sees applications in target identification, reprogramming protocol design and evaluation of aging interventions.
For now, VAC remains a preview rather than a validated digital counterpart of an aging biological system. Insilico has released an initial demonstration and plans to discuss development and validation of the platform at ARDD 2026 in Boston.
When the clock starts moving
The harder milestone will not be generating a plausible older or younger cell, but predicting how real biology moves between those states – particularly after an intervention nudges it off its expected course.
A virtual cell that can age is useful; one that can correctly predict what changes the direction of aging would be considerably more so.
