De Novo Protein & Chemical Design for Drug Discovery

De novo protein and chemical design for drug discovery

De novo design uses generative computational methods to create entirely new proteins and small molecules with desired properties, rather than screening existing libraries. Advances in deep generative models and structure prediction now allow researchers to design binders, biologics and novel chemical scaffolds tailored to a specific therapeutic target, dramatically expanding the accessible design space in drug discovery.

We combine modern generative AI with physics-based validation to propose, optimise and prioritise novel proteins and molecules that are focused, synthesisable and experimentally testable.

Our Services

• De novo protein and binder design (RFdiffusion, ProteinMPNN)
• Generative small-molecule design and optimisation
• Structure prediction and complex modelling (AlphaFold)
• Structure-based and ligand-based design
• ADMET and drug-likeness prediction
• Docking and molecular-dynamics validation
• Lead optimisation for potency and selectivity
• Candidate ranking and shortlisting for synthesis

This accelerates early discovery by delivering a focused set of high-quality, experimentally testable candidates.

What you receive

• A ranked, focused shortlist of designed candidates with the evidence behind each ranking
• Structural models of designs and their complexes with the target
• Predicted binding, stability, solubility and ADMET properties as applicable
• Molecular-dynamics validation showing which designs remain stable and bound
• Synthesis and expression considerations, so candidates are practical to make
• A written report and the full computational record

How designs are generated and filtered

Generative models can produce essentially unlimited candidates. The value is not in generating them but in filtering aggressively enough that what reaches the bench is worth the cost of making it.

• Target analysis: structure preparation, pocket or epitope definition, and assessment of whether the site is genuinely tractable
• Backbone generation with diffusion-based methods such as RFdiffusion for proteins, or generative chemistry for small molecules
• Sequence design onto generated backbones with ProteinMPNN or comparable methods
• In silico refolding to confirm each designed sequence is predicted to adopt the intended structure
• Complex modelling and docking against the target
• Molecular-dynamics simulation to test whether predicted interactions survive in explicit solvent
• Developability filtering: aggregation propensity, solubility, expression liability, synthetic accessibility

An honest account of what computation can promise

De novo design has produced genuine successes, including binders and enzymes with no natural counterpart. It also has a substantial failure rate: many designs do not express, do not fold as intended, or bind far more weakly than predicted.

We are explicit about confidence at each stage and design campaigns with that reality built in — producing a testable panel rather than a single candidate, and prioritising designs that are cheap to make and quick to falsify. Anyone promising computational certainty here is overselling.

Where this is used

• Binders and mini-binders against a defined therapeutic target
• Antibody and nanobody design and optimisation
• Novel chemical scaffolds where existing libraries have been exhausted
• Protein stabilisation and thermostability engineering
• Biosensors, diagnostics and research reagents