Pocket-aware generation
We generate novel, synthesizable chemistry purpose-built for a target's three-dimensional pocket — far beyond known libraries.
Multimodal AI · real-world physics · molecular mechanics
Multimodal AI for target identification and drug discovery.
Thea Biosciences unifies generative AI, physics-based simulation, and molecular-mechanics modeling into one discovery engine — built to find the right target and design the molecule that drugs it. Our first program, Titan, is a patent-pending ApoE4 structure-corrector series for Alzheimer's — now ready for synthesis, binding confirmation, and functional validation.
ApoE4
Our first program's target — the strongest genetic risk factor for late-onset Alzheimer's
55M+
people living with Alzheimer's & related dementias worldwide
Our focus
We specialize in genetically validated, structurally anchored, hard-to-drug targets — where an experimentally observed pocket exists, but no approved direct therapy does.
ApoE4 is exactly that target. Titan is our first program against it.
Why Titan is partner-ready
A diligence-ready preclinical asset — not a concept deck.
Everything a scientific or BD reviewer needs to evaluate Titan, in one place. The quantitative figures live in the data room; what they establish is below.
01
Target rationale
ApoE4 is the strongest common genetic risk driver for late-onset Alzheimer's. One copy raises risk several-fold; two copies raise it by roughly an order of magnitude. This is human genetics — high conviction, not hypothesis.
02
Structural anchor
Titan is designed against the experimentally observed ApoE4 small-molecule binding pocket — the published co-crystal structure family, not a computational guess.
6NCO / 6NCN co-crystalsHelix 1–4 interface1.7 Å resolution
03
Computational funnel
A disciplined, multi-stage selection — each candidate co-optimized for potency, selectivity, and safety inside a single search.
Generated libraryStructure-aware scoringCross-method rankingADMET / CNS filteringLead selection
04
Intellectual property
The lead series is novel composition of matter, protected by a filed provisional patent application. The platform methods are held as trade secrets and shared only under NDA.
05
Next experiments
A clear, fundable wet-lab plan to convert computational conviction into experimental evidence.
SynthesisSPR / ITC bindingCo-crystal attemptApoE4 cellular correction assayIn-vitro ADMEMicrosomal stabilityCaco-2 / MDCK-MDR1Plasma protein bindingEarly PK
The opportunity
The hardest targets in medicine are the ones that matter most.
Some of the most consequential disease biology has resisted drug discovery for decades — not for lack of interest, but because the chemistry is brutally hard. ApoE4 is the canonical example: overwhelming human genetic validation, and no approved drug that addresses it directly.
That gap — between a genetically validated target and zero direct therapies — is precisely where a platform built for hard, structurally anchored targets earns its keep.
Global prevalence
55M+
people living with dementia worldwide
Projected 2050
139M
cases worldwide by 2050
Annual cost of care
$1.3T+
global economic burden
Genetic risk
~15×
higher risk for two-copy ApoE4 carriers
Flagship program · Titan
A first program, designed and cross-validated in silico.
Titan is a series of novel small-molecule structure correctors for ApoE4 — the strongest genetic risk factor for late-onset Alzheimer's. Every candidate was generated, scored, and safety-profiled computationally, then cross-checked across independent methods.
- Experimentally observed pocket.Designed against the only published small-molecule binding site on ApoE4 (6NCO / 6NCN).
- Novel, patentable chemistry.A composition-of-matter genus protected by a filed provisional application.
- Safety designed in, not bolted on.Leads cleared multi-parameter ADMET and structural-alert screens.
The platform — second, but reusable
Multimodal AI, grounded in molecular mechanics and real-world physics.
Titan is the proof; the platform is the compounding asset. One integrated loop generates novel chemistry for a target's 3-D pocket, scores it against real-world physics — not docking heuristics — and anchors every candidate to the drug's mechanism of action. Potency, selectivity, and safety are optimized together, in a single search.
Physics-based scoring
Every molecule is evaluated with structure-aware, physics-grounded simulation against the observed binding site — then cross-checked by independent methods.
Safety-first optimization
Candidates are scored against a full multi-parameter ADMET profile — cardiac, hepatic, CNS exposure, structural alerts — before they ever advance.
Cross-isoform selectivity
We model selectivity across closely related isoforms — designing for the pathogenic form while sparing the benign.
How this maps to Titan: the platform spans the full arc — from identifying and validating targets to designing the molecules that drug them. For Titan we anchored on a target with overwhelming human-genetic validation (ApoE4) and its published co-crystal pocket, then let the engine do the hardest part — designing novel, safe, CNS-ready chemistry against it.
What is uniquely true about Thea
Technical differentiation — not marketing.
"AI for drug discovery" is crowded and mostly undifferentiated. Here is what is specifically, technically true about how Thea works — and why a well-funded competitor can't simply copy it.
/ 01
Physics, not pattern-matching
We rank candidates with structure-based, physics-grounded simulation against an experimentally observed pocket — modeling the real interaction, not a learned resemblance to known binders. Models trained on historical chemistry regress to the mean on genuinely novel scaffolds; physics does not.
/ 02
Anchored to the drug's mechanism
Our objective is the drug mechanic — restoring ApoE4's healthy conformation (structure correction) — not a bare affinity number. Every candidate maps to a therapeutic hypothesis, so a hit is a lead with a reason to work, not a docking artifact.
/ 03
One closed loop, co-optimized
Generation, physics scoring, cross-method validation, and a full ADMET / CNS safety profile run inside a single multi-objective search. Sequential pipelines discard the information each step needs — and quietly lose potent-but-toxic and safe-but-weak molecules at every hand-off.
/ 04
The integration is the moat
We design only against experimentally observed pockets and gate hard on safety. The edge isn't any single model — it's the proprietary stack (design rules, scoring formulation, optimizer tuning) for hard, structurally anchored targets, held as a trade secret. That integration — not "more AI" — is what's hard to replicate.
In short: anyone can wire a language model to a docking script. Thea's advantage is the disciplined fusion of real-world physics, mechanism-of-action design, and safety-first multi-objective optimization — proven end to end on a target most pipelines can't touch.
How we work
From anchored target to lead series — ready for the bench.
A disciplined, repeatable loop. Steps 01–04 are complete for Titan; step 05 is the partner-funded bridge to the lab.
01
DoneAnchor
Lock onto a validated target and an observed pocket.
02
DoneGenerate
Design novel, synthesizable chemistry for the pocket.
03
DoneScreen
Rank with physics-based, cross-validated scoring.
04
DoneOptimize
Co-optimize potency, selectivity, and safety together.
05
NextValidate (wet-lab)
Partner-funded experimental confirmation.
Steps 01–04 complete for Titan. We're raising the bridge to step 05 →
Why now
Three curves crossed at once.
01
A structural foothold
For the first time, an experimentally observed small-molecule pocket exists on ApoE4 — giving generative chemistry a real, physical anchor to design against.
02
Computational maturity
Generative models and physics-based simulation can now explore chemical space at a scale and fidelity that was impossible only a few years ago.
03
Unmet need at scale
Despite decades of effort and overwhelming human genetics, the most important neurodegeneration targets still have no disease-modifying therapy.
What we've established
Evidence, not assertion.
The quantitative results live in our data room. Here's what they establish — the figures themselves are shared under NDA.
Observed pocket
Designed against an experimentally observed binding pocket (6NCO / 6NCN).
Cross-validated potency
Top candidates confirmed by independent scoring methods.
Clean safety profile
Leads cleared multi-parameter ADMET & structural-alert screens.
Patent-pending
Novel composition of matter; provisional application filed.
The team
Built by founders who ship.
Thea Biosciences was founded by two builders who took a hard target from hypothesis to a patent-pending, cross-validated candidate series — using a computational pipeline they architected end to end.
JPJananthan Paramsothy, MD, MPH
Founder · Pipeline Architect
RSRamja Sritharan, PhD(c)
Co-Founder
White paper · v1.0
Designing ApoE4 Structure Correctors with a Multimodal ML Platform
The public edition covers the problem, our platform thesis, an anonymized selection funnel and lead-profile table, and an honest map of what is — and isn't — yet proven.
Research
The science, in the open. The recipe, protected.
We're glad to show our results and our reasoning. The architecture, training strategy, and design rules that produce them remain proprietary — available to qualified partners under NDA.
Partner with Thea Biosciences
Fund the bridge from in silico to in vivo.
We're seeking a wet-lab validation partner and bridge funding to confirm Titan at the bench — synthesis, binding, and a cell-based ApoE4 correction assay — and to extend the engine to the next target.
Or email us directly — partnerships@theabiosciences.com