Target identification
Genomics- and network-driven target discovery, druggability assessment, and evidence gathering to choose where to aim.
Research Area
Drug Discovery & AI
Discovery is a search problem across an impossibly large chemical space. Computation and AI are how you search it well—prioritising the few molecules worth making. We bring the tools: structure prediction, virtual screening, molecular dynamics, QSAR, and generative chemistry, for early-stage research.
Targets & structures
Virtual screening & docking
QSAR & generative AI
Research predictions · need validation
Illustrative sample output
library
filters
candidates
Overview
Searching chemical space, then handing the lab a shortlist
Drug-like chemical space is estimated to hold on the order of 1060 molecules—more than anyone can synthesise or assay. The value of computation is triage: predicting which structures fold and bind, which molecules are worth making, and which will fail on ADMET before a bench scientist spends a month on them. Done well, it turns an impossible search into a ranked, testable shortlist; done carelessly, it produces confident numbers that mean nothing.
We take the careful path. From target assessment and AlphaFold structure prediction to virtual screening, molecular dynamics, QSAR and generative models, and ADMET prediction, we apply established and modern methods—and return results documented for reproducibility. We are explicit about what a prediction is: a hypothesis that prioritises experiments, with its uncertainty stated. Computational results guide the lab; they do not replace experimental validation, and clinical safety and efficacy are established only in trials.
Applications
What we do in computational discovery
The analyses that narrow chemical space to testable candidates—from targets and structures to screening, generative design, and ADMET.
Protein-structure prediction
Structure prediction with AlphaFold and related models, plus binding-pocket identification and preparation for design.
Virtual screening & docking
Docking of large compound libraries against a target, with scoring and ranking to prioritise candidates.
Molecular dynamics & free energy
MD simulation of binding stability and free-energy calculations for more reliable affinity estimates.
QSAR & bioactivity ML
Machine-learning models for activity and property prediction from chemical structure, trained on curated assay data.
Generative chemistry
AI generation of novel, synthesizable molecules under your constraints—scaffolds, properties, and objectives.
ADMET & tox prediction
Prediction of absorption, distribution, metabolism, excretion, toxicity, and drug-likeness to flag liabilities early.
Drug repurposing
Signature- and network-based repurposing to connect existing drugs to new targets and indications for research.
The Pipeline
A representative discovery workflow
A transparent, staged funnel from a target and a library to a prioritised, testable shortlist—each filter documented and each score reported with its uncertainty.
Adapted to your project: structure-based vs. ligand-based, screening vs. generative, known target vs. discovery. We confirm the plan with you before any compute begins.
Target & data intake
We gather the target, available structures, compound libraries, and any assay data, and define the objective and constraints.
Sources: PDB · ChEMBL · your data
Structure preparation
Structures are predicted or curated, protonated, and prepared; binding pockets are identified and validated.
Tools: AlphaFold · ColabFold · pocket prep
Virtual screening & docking
The library is docked against the target, poses scored, and an initial ranking produced for triage.
Tools: AutoDock Vina · DiffDock · Smina
ML scoring & ADMET filtering
QSAR and ML models refine activity estimates, and ADMET and drug-likeness filters remove liabilities.
Tools: DeepChem · RDKit · ADMET-AI
MD & free-energy refinement
Top candidates are checked with molecular dynamics and free-energy calculations for binding stability.
Tools: GROMACS · OpenMM · FEP
Candidate prioritization
Evidence is combined into a ranked, annotated shortlist with scores, predicted properties, and stated uncertainty.
Output: ranked candidates · rationale
Reporting & handoff
Results become clear figures, tables, and reproducible methods—framed as hypotheses for your experimental validation.
Tools: reports · figures · versioned manifest
Tools & Technologies
Established, peer-reviewed tools—matched to your data
We select from the field's standard toolkit rather than forcing every dataset through one pipeline. A representative set of what we work with:
Structure prediction
AlphaFold2/3
ColabFold
ESMFold
RoseTTAFold
Docking & screening
AutoDock Vina
DiffDock
Smina
rDock
Molecular dynamics
GROMACS
OpenMM
AMBER
NAMD
Cheminformatics
RDKit
OpenBabel
DataMol
Pmapper
ML & generative
DeepChem
PyTorch
DGL-LifeSci
REINVENT
ADMET & toxicity
ADMET-AI
SwissADME
pkCSM
DeepPurpose
Free energy
alchemlyb
PMX
MM/PBSA
FEP
Databases
ChEMBL
PubChem
PDB
DrugBank
Key Databases
The chemistry & target resources we draw on
Discovery is powered by curated structure, bioactivity, and compound databases. We build on the community's authoritative, versioned resources.
- PDB
- The Protein Data Bank of experimental 3D structures.
- ChEMBL
- Curated bioactivity data linking compounds to targets.
- PubChem
- Large open repository of chemical structures and assays.
- DrugBank
- Detailed drug, target, and mechanism information.
- BindingDB
- Measured protein–ligand binding affinities.
- ZINC
- Purchasable compound libraries for virtual screening.
- Open Targets
- Target–disease evidence for target selection.
- UniProt
- Protein sequence and functional annotation.
- ChEBI
- Ontology of chemical entities of biological interest.
Choosing an Approach
Structure-based vs. ligand-based vs. generative
Different starting points call for different methods. A quick orientation; we will help you match it to your project.
| Dimension | Structure-based | Ligand-based | Generative |
|---|---|---|---|
| Needs | Target 3D structure | Known active molecules | Objectives & constraints |
| Uses | Docking & MD in the pocket | QSAR & similarity models | AI molecule generation |
| Output | Ranked docked poses | Predicted-active analogs | Novel candidate molecules |
| Explores novelty | Within the library | Around known chemotypes | Beyond known chemotypes |
| Best suited to | Well-characterised targets | Rich assay history | New chemical matter |
What You Receive
A complete, documented deliverable
Not just a docking score dumped in a folder—a defensible, ranked shortlist your chemists can act on, with every assumption documented.
- Ranked candidate table with docking, ML, and ADMET scores
- Predicted or prepared target structures & binding-site analysis
- Docked poses & interaction diagrams for top candidates
- MD / free-energy results, where run, with stability metrics
- ADMET & drug-likeness flags per candidate
- Generated molecules with property profiles, where applicable
- Reproducible methods with model versions & stated uncertainty
Built for reproducibility, not just a result
Computational discovery follows documented best practices—validated pockets, physically sane docking and MD setups, models trained and tested on held-out data with reported metrics, and every model and library version pinned—so a top-ranked candidate is a defensible hypothesis rather than a lucky score. We are explicit about the limits: docking scores, ADMET predictions, and ML activity models carry real uncertainty and are hypotheses that prioritise experiments, not guarantees of potency, safety, or efficacy. Molecules must be validated in the lab, and clinical safety and efficacy are established only through trials. We accelerate discovery; we do not replace it.
The practical payoff: your methods section writes itself, a reviewer can re-run the analysis, and a result from today can be reproduced a year from now. We will also tell you honestly when a design or sample size won't support the conclusion you're after.
FAQ
Drug discovery & AI questions
What discovery and medicinal-chemistry teams most often ask before starting.
It depends on the task: a protein target (sequence or structure), a compound library or set of molecules (SMILES), and any assay or bioactivity data you have. We can predict structures where none exist and pull compounds and activity data from public databases like ChEMBL and PubChem.
Yes. We predict structures with AlphaFold and related models, identify and prepare binding pockets, and model protein-ligand and protein-protein interactions for structure-based design.
Yes. We dock large compound libraries against a target, score and rank candidates, and combine docking with ML and ADMET filters to prioritise a shortlist for experimental testing.
Yes. We use generative models to propose novel molecules under your constraints, and predict ADMET and drug-likeness properties to flag liabilities early. These are computational predictions to guide chemistry, not measured values.
No. Docking scores, ADMET predictions, and ML activity models are hypotheses that prioritise candidates for experiments — they are not guarantees of potency, safety, or efficacy. Real molecules must be validated in the lab, and clinical safety and efficacy require trials. We accelerate discovery; we don't replace it.
Drug discovery is a research area we support through our core services — chiefly Proteomics & Structural Analysis, Custom Pipelines & Software, and Biostatistics & Visualization — applied to targets and compounds. This page shows how they come together.
Related
Services that power discovery work
Have a target or a library to screen?
Tell us your organism, data type, and question—we'll scope it honestly, including if a different design would serve you better.