Cloud
Run with us
- Fast setup, managed hardware
- Pay-per-run, no infrastructure burden
- Same Composer, same models
- Scaling handled automatically
Best for teams who want to move fast and don't have strict data residency requirements.
Book an audit
AI infrastructure for biotech
For early-stage biotech teams and wet labs
Biological AI your scientists can actually run.
We design and ship model-driven workflows for early-stage biotech teams and wet labs: antibody triage, variant interpretation, target-to-hit. Run them in the cloud, or keep everything local on your hardware. Your scientists run them from a form, not a notebook.
Free 30-minute call. Bring one bottleneck, leave with a concrete plan.
- Biological AI models ready to run
- 17
- Ready-made pipeline templates
- 10
- Notebooks your scientists need to touch
- 0
Runs models your team already reads about
- OpenFold3
- AlphaFold2
- Boltz-2
- ESM-2
- Evo2
- RFdiffusion
- ProteinMPNN
- DiffDock
- AlphaMissense
- DNABERT-2
- ADMET-AI
The gap
The models are public. The workflow around them isn't.
-
01
Notebooks only one person can run
The model code lives in one postdoc's notebook. When they're busy or gone, the analysis stops.
-
02
Weeks of setup before the first result
GPUs, containers, API keys and file formats eat the time that should go to the science.
-
03
Outputs nobody else can read
Raw scores and PDB files don't tell a bench scientist what to test next.
Novopharos turns each of these into a tool your whole team can run, with outputs framed around the decision you're actually making.
Product tour
See the platform do the work.
Four things your team gets on day one.
Chain models on a canvas. Run them with one click.
Drop structure prediction, scoring, filtering and literature search onto a canvas. Links are checked as you draw, so a broken pipeline never reaches a GPU.
- 10 ready-made templates to start from
- Invalid connections caught instantly
- Save pipelines and re-run them any time
17 models. One form. No environment to set up.
Paste a sequence, pick a model, press run. OpenFold3, Boltz-2, ESM-2, Evo2 and the rest of the hub run on managed infrastructure with sensible defaults.
- Structure, protein LM, genomics, design and small-molecule models
- Example inputs to start from
- Pay-per-run credits instead of idle GPU bills
Results laid out for the next decision.
Structures with per-residue confidence, ranked mutation tables and confidence-colored sequences, so a bench scientist can see what to test next.
- 3D structure viewer with pLDDT confidence
- Every substitution scored and ranked
- Download PDB and FASTA for downstream tools
Paste a lab page. Get a pipeline built for its science.
Novopharos reads a lab's current work, recent papers and methods, then proposes a pipeline from our models, with the reasoning behind every step.
- Works from a lab URL or a researcher's name
- Each step linked to the papers that justify it
- Edit it on the canvas before you run
Your choice: cloud-native or fully local.
Deploy the Composer and models on our managed infrastructure, or we'll containerize everything for your own hardware—data never leaves your network.
Local
Run on your infrastructure
- All data stays on your hardware
- Full ownership and control
- Ollama-based inference on your GPU/CPU
- We design, containerize, and deploy
Best for labs with compliance requirements or sensitive research that can't leave your network.
Book an audit
Workflow case studies
Workflows we build for teams like yours.
Three example engagements, each built in the Novopharos Composer. The screens below are from the product.
Antibody variant triage
For antibody startups and CROs screening variant panels
The bottleneck
Hundreds of variants per round and wet-lab capacity for ten. Triage happens in spreadsheets, and structure checks run one sequence at a time.
The workflow
- Import antibody sequences
- ESM-2 fitness scoring
- OpenFold3 structure prediction
- Literature prior-art check
- Germline identity filter
- Export the top 10 candidates
What your team gets
- A ranked shortlist with structures and confidence scores
- Known liabilities flagged from the literature
- The same pipeline, re-run every round with one click
Rare variant interpretation
For clinical genomics and rare-disease research teams
The bottleneck
Every VCF means hours of lookups across pathogenicity databases, sequence models and papers before the team can discuss a single variant.
The workflow
- Load the VCF and parse variants
- AlphaMissense pathogenicity
- DNABERT-2 sequence impact
- Clinical literature search
- Results summary
- PDF report for review
What your team gets
- Every variant annotated with pathogenicity and sequence-impact scores
- Supporting literature attached to each call
- A report the whole team can review together
Target-to-hit
For small-molecule drug discovery startups
The bottleneck
Going from a target gene to docked, drug-like hits means stitching together structure prediction, pocket finding, docking and ADMET tools that don't talk to each other.
The workflow
- UniProt sequence fetch
- OpenFold3 target structure
- P2Rank pocket detection
- DiffDock blind docking
- AutoDock Vina rescoring
- ADMET-AI liability filter, then ranked hits
What your team gets
- Druggable pockets ranked on your target
- Hits docked, rescored and screened for ADMET liabilities
- A ranked list your chemists can act on
Have a different workflow in mind? Most engagements start with one we haven't built yet.
Talk through your workflow
What we do
An engineering partner that understands lab reality.
Embedded AI engineering
We scope, build and ship internal tools for teams that need ML capability before they can hire a computational biology group.
Scientific workflow software
We turn messy bench-to-analysis processes into reliable interfaces and orchestration your team can operate without us.
Biological model operations
Parameter surfaces, run controls, scoring views and handoff into downstream pipelines, so foundation models work in practice.
How we work
From bottleneck to working tool in weeks, not quarters.
-
Week 1
Workflow audit
We map the bottleneck, the data you have, and the decision the output has to support. You get a written workflow plan.
-
Weeks 2–4
First workflow live
We build it on Novopharos, validate it on your data, and hand your scientists a tool they run themselves.
-
Then
Expand into a platform
Workflows that prove themselves become reusable infrastructure across programs, models and teams.
FAQ
Questions teams ask before the first call.
Where does our data go?
Your sequences and results stay in your Novopharos workspace. Models run on managed endpoints such as NVIDIA NIM, or on infrastructure we deploy for you when data can't leave your environment. We're happy to sign an NDA before seeing anything sensitive.
Who owns the results?
You do: your data, your results, and the pipelines built on them.
Which models do you support?
17 are ready to run today: structure prediction (OpenFold3, Boltz-2, OpenFold2, AlphaFold2, MSA Search, P2Rank), protein language models (ESM-2, ESM-1b), genomics (Evo2, DNABERT-2), protein design (RFdiffusion, ProteinMPNN) and small molecules (DiffDock, AutoDock Vina, ADMET-AI, GenMol, MolMIM). More are on the way, and if you need one we don't have, we add it.
Do we need GPUs or an ML engineer?
No. Models run on managed infrastructure, and your scientists work through forms and the Composer. We handle the engineering.
Will it fit our existing tools?
Results export as standard files like PDB and FASTA. During an engagement we connect to where your data already lives, whether that's an ELN, a LIMS or a cloud bucket.
What does it cost?
The audit call is free. Pilots are fixed-scope and quoted once we understand the workflow. Platform usage is billed as credits, so you only pay for the runs you make.
Book a call
Bring one bottleneck. Leave with a plan.
On the call
- We map the workflow that's slowing your team down
- We tell you honestly whether AI will help
- You leave with a concrete next step, whether or not we work together