Drug discovery keeps falling into the same trap
Teams design against a single tidy structure. Real proteins behave like dynamic machines. Loops shift. Domains breathe. Pockets open and close. The most druggable state may exist only briefly.
AlphaFold made static structure prediction fast and accessible. That was a major leap forward. But a still structure cannot answer:
When does a hidden pocket open?
How long does it remain accessible?
What structural rearrangement is required to reach it?
Which states are stable and which are rare?
Designing against a static structure often means optimizing against a conformation that does not dominate under real biological conditions.
The Trusted Tool and Why It Does Not Scale
Brute force molecular dynamics remains the gold standard for resolving protein motion.
Molecular dynamics is an excellent judge. It is not an efficient explorer.
The Compounding Cost of State Finding
The cost compounds because exploration is never one protein one run.
A typical effort to identify important states requires:
- 100 to 1,000 GPU days
- 2,400 to 24,000 GPU hours
At common H100 class pricing:
- About 6.88 USD per GPU hour on AWS p5.48xlarge
- About 12.29 USD per GPU hour on Azure ND96isr H100 v5
Compute alone lands between 16.5K and 295K USD per protein. That is before reruns, alternate conditions, replicas, or specialist labor.
Now multiply by real portfolio complexity. Twenty proteins across targets, off targets, variants, and constructs can approach 5.9M USD in exploration compute. Ten programs across a pipeline can reach roughly 59M USD at the high end.
That is not candidate optimization. That is the cost of discovering which states matter before optimization even begins. This is the structural tax MDForce™ is designed to reduce.
The Market Landscape
The ecosystem today falls into three categories.
Better Still Photos
Structure prediction platforms improve static models
Valuable progress, but still snapshots.
Faster Calculators
Some companies focus on accelerating simulation workflows
NVIDIA enables AI driven potentials inside mainstream simulation engines. Acellera advances high throughput molecular dynamics and adaptive sampling and is releasing learned potentials such as AceFF.
This wave is important. However, most approaches still advance step by step through time. Rare gating events and pocket openings can remain slow to appear. They make simulation faster. They do not change the exploration strategy.
Full Stack Drug Factories
Some platforms aim to own the entire pipeline
Isomorphic Labs describes an in silico drug design engine used across programs and partnerships.
Recursion positions its platform as an end to end system combining large scale data, algorithms, and lab feedback. These approaches can be powerful, but they often require deep workflow replacement and introduce higher levels of lock in.
Where Vareon MDForce™ Fits
Vareon occupies a distinct layer.
- We are not a faster calculator.
- We are not a full stack replacement.
We are the motion infrastructure
If AlphaFold is a still image and brute force molecular dynamics is the expensive test flight, MDForce™ is the flight simulator.
Teams explore thousands of realistic motion trajectories efficiently and advance only the strongest candidates to high fidelity validation.
How MDForce™ Works
Step 1
Make the Starting Point Reliable
Exploration quality depends on input quality. We standardize predicted models, experimental structures, and legacy inputs into consistent and dependable starting states. Reducing structural inconsistency prevents downstream compute waste.
Step 2
Make Motion Cheap Enough to Use Early
Instead of spending weeks on brute force simulation just to discover relevant states, MDForce™ generates realistic conformations and transition paths in hours to days.
Teams can:
- Explore thousands of plausible states
- Identify transient pocket openings
- Rank stable versus rare conformations
- Shortlist motion aware candidates
Heavy molecular dynamics is then reserved for confirmation.
Many AI plus molecular dynamics approaches improve simulation efficiency. MDForce™ changes how exploration itself happens.
Step 3
Connect Motion to High Stakes Decisions
Late surprises destroy value. Programs can appear promising until a pocket collapses, a loop shifts, or a candidate binds only to a rare state. Many tools provide a single score.
MDForce™ provides motion aware stress testing:
- Which conformations are robust across conditions
- Which states are transient
- Which designs depend on fragile geometries
This reduces optimization against unstable targets.
Step 4
Extend to Multi Body Reality
MDForce™ expands from single protein motion toward multi body interaction layers.
We do not claim to simulate entire cells. We provide a molecularly faithful motion layer that feeds grounded constraints into higher order system models so what is built on top behaves realistically.
Built for Pharma Operations
Adoption inside pharma requires operational reliability. The same input must produce the same output. Runs must be replayable. Results must be traceable and auditable.
Each execution records:
Dataset Identifiers
Preprocessing Steps
Random Seeds
Model Versions
Evaluation Outcomes
Deployments reference immutable artifact digests with signed configurations
Motion is governed by consistent executable rules rather than stitched frame sequences. This supports enterprise deployment, regulatory defensibility, and reproducible science.
The moat compounds over time
Not because we thought of motion first. Because we become the daily motion layer inside design workflows. This leads to:
- Reusable target specific state libraries
- Workflow embedment inside pharma environments
- Accumulating archives of motion patterns and outcomes
- Cross target acceleration over time
Each new protein increases performance on the next. Built for scale through batched GPU execution and distributed evaluation across large portfolios.
This is infrastructure, not boutique simulation
We begin realistically
Public molecular motion datasets including Folding at home outputs provide breadth. We standardize them into training grade motion datasets.
The End State
Vareon MDForce™ turns protein motion into a reusable everyday capability inside the design loop.
High fidelity molecular dynamics remains the final judge. But it is no longer the primary explorer.
We reduce the repeated portfolio wide tax of running weeks of simulation just to discover which state should have been used for design.
Savings scale with every additional protein. With every additional program. With every portfolio expansion.
This is a structural shift in how motion is used in drug discovery.
