Our proprietary AI models, ALScreen, ALGen and ChaRNABERT among them, work hand in hand with PELE®, our physics-based core, and with TPD-Suite®, GluePELE, PELETac and XNAHub®. Access them through our discovery projects, as licensed software or on our high-performance cloud.
AI Platform
Want AI built around your discovery program?
Our AI department builds proprietary pipelines that take a program from target hypothesis to prioritized molecules, and adapts them to each partner’s target, data and constraints. Our virtual screening, hit generation and ADMET models have been validated experimentally.
- Target discovery with drugESM and Yggdrasil, to prioritize biological targets
- Druggability with PickPocket, to identify pockets and assess tractability
- Virtual screening with ALScreen and Tensor-DTI, to rank ultra-large libraries
- Hit generation with ALGen, to design novel chemotypes
- ADMET prediction with state-of-the-art models, to flag developability liabilities early
ALScreen
Need to search billions of compounds without docking every one?
ALScreen, our hybrid AI and molecular modeling workflow, docks a representative subset, trains a surrogate model on the scores and runs it across the full library. Active learning then focuses each new round on better compounds.
- Screens up to 200 billion molecules in a few weeks
- 5 billion compounds scored in about 6 hours on CPUs alone (~6k core-hours), versus an estimated 2-5 years of exhaustive docking
- Experimentally validated on three hard-to-drug interleukin PPI targets: 65 billion make-on-demand compounds screened, binders confirmed by synthesis and SPR (best at 29.6 µM)
- Adaptable to your target, chemical space, project constraints and docking strategy
ALGen
Looking for novel chemotypes, not just new analogs?
ALGen merges generative AI with a physics-based active learning loop. It generates, filters, evaluates, selects and retrains, round after round, moving toward the chemical profile each project needs. Combined with ALScreen, it screens existing chemical space and creates new space, converging on testable hit packages.
- 8 of 9 synthesized CDK2 designs active, best inhibitor at 71 nM
- Novel, synthesizable chemotypes, checked against 76 billion make-on-demand compounds
- Ranked compounds and novel designs delivered against client-specific criteria
- Published in Communications Chemistry (2025)
ChaRNABERT
Working on RNA targets?
ChaRNABERT is our foundational RNA language model. It turns RNA sequences into efficient representations that support structure and binding-site questions, and it already runs in client RNA pipelines.
- Efficient RNA representations for downstream structure tasks
- Binding-site related tasks for RNA target analysis
- Applied in client RNA pipelines, beyond academic benchmarks
- Peer-reviewed at ICLR 2025 (LMRL Workshop)
PELE® and Nostrum Suite®
Need physics you can trust?
PELE (Protein Energy Landscape Exploration), our core Monte Carlo algorithm, explores the flexibility of both ligand and protein at a low computational cost. Nostrum Suite® builds on PELE, integrating physics-based modeling with AI to accelerate drug discovery and molecular engineering.
- Binding site prediction and protein surface exploration
- Docking pose refinement and ranking
- Ligand migration and exit-path modeling
- The engine behind TPD-Suite®, GluePELE and PELETac
- Protein-protein docking, plus cutting-edge proprietary and open-source molecular modeling and simulation software
Software Licensing
Want to run our software in-house?
We license Nostrum Suite® to third parties, giving your team direct access to the same workflows we use in our own projects.
- Premium: for intensive use, or to run the software on your own premises
- Cloud: unlimited access to Nostrum Suite® on a private HPC cluster set up by Nostrum Biodiscovery
- Web app: selected workflows in your browser, at app.nostrumbiodiscovery.com
Custom Developments
Need a workflow built around your process?
We design automated workflows from scratch or adapt our existing solutions to your requirements, including graphical interfaces to prepare inputs, run simulations and analyze results. Our desktop app is built on Horus, a multi-platform GUI that runs locally or as a central server for team collaboration.
- End-to-end automation, for example from 1D/2D protein and ligand data to predicted structure, docking and binding-mode analysis
- Custom applications and graphical interfaces
- Horus blocks that link into shareable workflows through an accessible API
- Local or server deployment for collaborative teams
Infrastructure
Need compute at scale?
Our infrastructure combines an on-demand cloud cluster with our own on-premise HPC cluster. Our software is compiled and optimized at the microarchitecture level, so it runs as fast as possible at the lowest cost.
- Cloud cluster: on-demand access to thousands of CPU cores and tens of GPUs
- On-premise HPC cluster: 10+ nodes with hundreds of CPU cores and tens of GPUs
- Private cloud instances, preinstalled with state-of-the-art proprietary and open-source molecular modeling and AI software
