Building UHBSE: a clinical simulator for patient-specific biological futures before intervention. Grounded in 7 years of real-world longitudinal patient data.
Wilmington, DE
- Tweets 285
- Following 108
- Followers 26
- Likes 149
Pinned Tweet
Simulate before you intervene.
UHBSE is building a clinical simulator to compare patient-specific biological futures before intervention. Grounded in seven years of real-world longitudinal patient data.
For design partners and investors: uhbse.tech
18
The paired chromatin and expression measurements make this a valuable mechanistic map. For the simulator we’re building at UHBSE, a key question is how such mechanisms relate to trajectories over time. Could this atlas be connected to repeated clinical measurements?
Our single-nucleus immune multiome atlas is now out in @Nature! 🧬
10M PBMCs from 1,108 @FinnGen_FI donors recruited by @FRCBSresearch, profiled at @broadinstitute KCO with chromatin accessibility and gene expression in the same nuclei, to trace how disease variants act 🧵👇
7
More measurements do not guarantee more meaning. UHBSE is building a clinical simulator to compare possible patient-specific futures before intervention, grounded in seven years of real-world longitudinal histories. Calibration and intervention effects need separate validation.
Short 🧵 on my new @WSJOpinion op-ed about measurement, meaning, and AI in health:
1/ Across personal health/wearables and pharma R&D, we can collect increasingly rich proxy measurements, yet often these can’t sustain the precision insights we want -- and increasingly expect.
10
Individualized care needs a view of how a person changes over time, not only a better-organized record of the past. The difficult test for any clinical simulator is whether its trajectories stay calibrated across care settings and patient groups.
I recently shared Mayo Clinic’s perspective on why the future of healthcare requires us to rethink its underlying data architecture and physical infrastructure to deliver the precise, individualized care patients want and deserve. Read more: investmentreports.co/intervi…
7
37,000 AI agents can read trial histories. A clinical simulator asks a different question: what could happen to one person over time, with or without an intervention? The Virtual Biotech study makes that distinction worth discussing. doi.org/10.1126/science.aeg6…
Biohub Investigator @james_y_zou built a “virtual biotech”—37K AI agents that analyzed 57K clinical trials and found clues to what makes drugs succeed. @nytimes traces its roots to a collaboration sparked by our scientist John Pak: bit.ly/4hLrpnm
4
29
Patient-level simulation needs real longitudinal histories, a natural-history baseline, treatment and observation timing, and uncertainty about where predictions apply. The test is whether predictions hold on patients and settings the model has not seen.
1
5
The study used agents to analyze more than 55,000 past trials and connect outcomes with molecular evidence. It found an association between cell-type-specific targets and trial progression. Useful for discovery; not proof of a causal treatment effect for a future patient.
5
UHBSE is moving from thesis to verified infrastructure.
Our auditable runtime reconstructs time-valid biological states and produces source-specific 90-day natural-history distributions.
Next: design-partner validation toward intervention modelling.
uhbse.tech
1
2
26
Sergei Narykov | UHBSE retweeted
I feel like this paper didn't get enough airtime when the preprint came out. Happy to see it published now! Great experiment (may annoy your local medchemist) nature.com/articles/s41586-0…
1
7
21
2,649
UHBSE is not built around one snapshot or one answer.
Its edge: seven years of real longitudinal clinical data, time-valid states, multiple possible trajectories, explicit uncertainty and traceable evidence.
Seeking biopharma/CRO design partner.
uhbse.tech
1
28
UHBSE is a research layer for comparing possible human biological trajectories before an intervention’s effect is treated as known.The intended product is an auditable human biology simulator. Today, it remains a research prototype.
1
1
64
What is not yet established: patient-specific intervention response, causal treatment effects, clinical validation or production readiness.
UHBSE does not recommend treatments or replace clinical trials. When evidence is insufficient, it must abstain.
1
5
We are looking for one biopharma or CRO design partner with a bounded longitudinal research question and access to independent data.
Together, we can freeze the protocol, test it on unseen data and produce an auditable validation package.
uhbse.tech
11
UHBSE is a human biology simulator built on real longitudinal clinical data. It is designed to estimate how a person’s biological state may change over time under different interventions and conditions and return a range of possible outcomes, not one certain answer.
8
30
Follow @uhbse_engine to see what survives validation, what fails, and how this becomes a useful research product.
We are looking for biopharma, CRO and health-data partners for a bounded design project.
Technical demonstration: uhbse.tech
12