Companies in Apheris-powered networks
Outperform existing models. Customize to your programs. Run inside your environment.
Companies in Apheris-powered networks
In collaboration with our partners
Data connected and results delivered
>25k
Structures
proprietary co-folding structures
>100m
Binding
proprietary binding datapoints
>900k
ADMET
ADME and tox measurements across 700 tasks
10,000
Antibody developability
ML-ready dataset in build with Ginkgo
52.1%
of predictions reach experimental-grade interface accuracy (PL-lDDT ≥ 0.8), vs 38.4% for the average public model
46.8%
of predictions reach an accurate ligand pose (bisyRMSD ≤ 2 Å), vs 32.2% for the average public model
28
ADMET endpoints pushed past the usability bar that members could not model reliably alone
OUR NETWORKS
AI Structural Biology (AISB) Network
Structure and binding
For virtual screening and lead optimization; Antibody-antigen co-folding for affinity maturation and protein design.
ADMET Network
Small molecules
Improve small-molecule ADMET predictions on novel chemistry, so teams triage series earlier and use experimental capacity better.
Antibody Developability Network
Large molecules
Advance antibody R&D with federated AI training on purpose-built datasets, in collaboration with Ginkgo Datapoints.
PROOF FROM A DATA NETWORK
Ahead of every public model tested, on both measures.
Axes: fraction of structures with PL-lDDT ≥ 0.8 and fraction with bisyRMSD ≤ 2 Å, on 1,056 held-out private structures from five pharma partners, ranked selection.
The results shown here are from AISB-1, evaluated on the largest proprietary industry benchmark of data no model has ever seen before, split to prevent data leakage. For a discovery team, more reliable predictions mean fewer wasted make-test cycles and more confident go/no-go decisions.
HOW IT WORKS
Run models in your own environment
Continuously customize models to your own data
Powerful workflows to embed models in drug programs
WHY APHERIS
Why leading pharma build with Apheris
Neutral by design
Apheris operates the networks and runs no drug programs of its own. Pharma build with us because we don’t compete with them.
It’s the data, not the architecture
We’re model-agnostic: we take the best models, open-source or your own, and lift them on your targets with data no public dataset can match, exactly where public models fall short.
Adopted through Foundry
The advantage reaches your programs through Foundry, the platform your teams run in-house to put these models to work across the DMTA cycle.
Your data never leaves your environment
Every computation runs inside your own environment. Your raw data is never pooled, copied, or transferred.
FAQ
Frequently asked questions
NEWSLETTER
A quarterly briefing on operationalizing and customizing co-folding, ADMET and antibody developability models. Written by Robin Röhm, co-founder and CEO at Apheris.
