Overview
What this category covers
This category is the machine-learning engine behind small molecule design: structure-conditioned generative models that build molecules into a pocket, and reinforcement learning that optimises them against several objectives at once. The modelling itself is the easy half. The work that determines whether the output is useful is reward design, filtering, and confirming every delivered molecule with physics-based methods.
What we deliver
▸Three-dimensional generative design for de novo molecules, fragment growing and linking
▸Scaffold hopping under shape, pharmacophore and substructure constraints
▸Multi-objective reinforcement learning with Pareto trade-off analysis
▸Reward function design, documentation and validation against your priorities
▸Physics-based confirmation of every delivered molecule before it reaches your chemists
Who this is for
▸Programmes wanting generative design as a capability rather than a one-off study
▸Teams optimising a series against several competing properties
▸Groups with internal predictive models they want incorporated into a design loop
▸Projects generating focused libraries around a validated scaffold
How engagements in this category are run
Platform engagements begin with the scoring or reward definition, because that is what determines whether the output is useful. We validate the models against your existing data where it exists, run generation or optimisation in rounds, and confirm everything delivered with physics-based methods. Where you can return assay results between rounds, models are retrained on them, which is consistently more effective than optimising against predictions alone.
Services
2 computational services in this category
F1
Generative Molecular Design (Diffusion Models)
State-of-the-art structure-conditioned generative AI targeting specific binding sites. 3D-aware molecule generation with drug-like constraints.
Learn MoreF2
Reinforcement Learning for Molecule Optimization
Reinforcement-learning-guided iterative optimisation against several objectives at once. Produces focused compound sets with the potency, property and synthesisability trade-offs made explic…
Learn MoreIn Practice
What a project in this category looks like
Three shapes we are asked for most often. Scope is always set after a feasibility review, so timelines below are typical rather than quoted.
Standing generative capability for a live programme
The situation
A discovery team wants generative design available continuously rather than as a one-off study.
What we do
Scoring and constraint definition agreed up front, generation run in rounds against the current structure and SAR, physics-based confirmation of everything delivered, and model updates as assay results return.
What you receive
A design set each round with poses, property profiles and confirmation results, plus the record of how the design space moved between rounds.
Typical timeline: Rounds of 3–4 weeks
Rescuing a potent series that fails on properties
The situation
Potency is good but solubility, permeability or metabolic stability blocks progression, and single-property fixes keep costing activity.
What we do
A multi-objective reward is constructed with you and pressure-tested, reinforcement-learning optimisation runs under scaffold constraints, and results are delivered as a trade-off surface rather than a single ranking.
What you receive
A Pareto front across your objectives with the molecules at each trade-off point, so the compromise is chosen deliberately rather than inherited.
Typical timeline: 5–8 weeks
Focused library generation around a confirmed scaffold
The situation
The core is fixed and the team wants systematic exploration of substitution rather than ad hoc analogue design.
What we do
Reaction-aware generation under scaffold constraints, property and liability filtering, diversity analysis and confirmation docking of the selected set.
What you receive
A focused set with predicted profiles, diversity statistics and synthesis routes, structured so the results will produce interpretable structure–activity relationships.
Typical timeline: 4–6 weeks
Questions
Asked before we scope this work
What data do we need before starting?
A target or a predictive model and a starting scaffold are enough. Measured series data improves the scoring function substantially, and without it the work leans more on generic predictors — which we will say plainly.
How do you stop the model gaming the score?
Output is inspected for the pathological structures these methods find when a reward is exploitable, and the reward is revised when that behaviour appears. It is the main reason reward design is treated as the core task.
Can you incorporate our internal models?
Yes. Internal predictors can be built into the scoring function, which usually makes the output considerably more relevant to your programme than public models alone.
What Published Work Reports
The results below were published by independent groups and are cited to show what these computational approaches have achieved in the literature. They are not ICDMO project results, and outcomes vary substantially between targets, datasets and programmes. We do not promise comparable results for any particular project.
A generative-AI-designed inhibitor reported positive Phase 2a lung function data
A programme in which both the target and the molecule were identified using AI methods reached a randomised Phase 2a trial in idiopathic pulmonary fibrosis. Across 71 patients treated for 12 weeks, the 60 mg once-daily arm showed a mean forced vital capacity change of +98.4 mL against −20.3 mL in the placebo arm, with a manageable safety profile. It remains an early-phase result in a small trial, but it is the first published clinical proof of concept for an AI-discovered target and AI-designed compound.
Nature Medicine, 2025 ↗More
Other service categories
Target Identification & Validation5Virtual Screening (Small Molecule)3Small Molecule Design & Optimization2ADMET & Drug-Likeness Prediction4Molecular Dynamics & Simulation3Structure-Based Drug Design (SBDD)3Bioinformatics & Multi-Omics Analysis11Antibody Design & Engineering (AI)5Peptide & Cyclic Peptide Drug Design4Protein Design & Engineering (AI)4ADC & Bioconjugate Design2Compound & Biological Library Design3
Bring us the scientific question, not a service code
Share your target, dataset, sequences or objective. We will recommend a scoped computational plan, the expected deliverables, a timeline and the validation options.
Request a Feasibility Review