In-Silico Evaluation Suite
Test the molecule before you make the molecule
The In-Silico Evaluation Suite is a complete computational workbench - docking, molecular dynamics, binding free energy, ADMET, QSAR, generative design and multi-objective optimisation - in one environment, on one data model, with one audit trail. Twenty-six modules that would otherwise be twenty-six separate tools and six file-format conversions.
- 26
- Modules in one environment
- 10
- Capability areas, one data model
- 0
- File-format conversions between stages
- 1
- Audit trail across the whole campaign
Every module reads and writes the same project record - compounds, targets, parameters, results - so a pose that came out of docking is the same object the dynamics run, the ADMET profile and the report refer to.
Key features
Ten capability areas, twenty-six modules, one project record
Open a capability to see what it does and the screen it does it in. Nothing here is a separate application - each one is reading the same compounds, the same targets and the same parameter history as the rest.
01Compound library management
Chemistry arrives in whatever format it arrives in. The library standardises it once, on ingest, so every downstream module is looking at the same molecule.
- Import from SDF, SMILES, MOL2, CSV and direct database pull
- Automatic standardisation: salt stripping, tautomer and protonation-state handling, charge assignment
- Descriptor and fingerprint calculation on ingest
- Duplicate detection, scaffold clustering and diversity analysis
- Project-scoped collections with full version history
In-Silico Suite · Compound libraryGrid view with structures, computed properties and cluster filters. 02Target & protein management
A structure is not ready to dock into because it came out of the PDB. The preparation pipeline is explicit, recorded and repeatable.
- Structure retrieval from the Protein Data Bank and AlphaFold
- Preparation pipeline: missing-residue repair, protonation, minimisation
- Binding-site detection and characterisation
- Homology model building where no experimental structure exists
- Pocket comparison across related targets
In-Silico Suite · Protein preparationDetected binding pockets ranked by druggability score. 03Molecular docking engine
Four docking modes, several scoring functions, and pose clustering that tells you whether the top pose is a family or an outlier.
- Rigid, flexible-ligand, induced-fit and covalent docking modes
- Ensemble docking across multiple receptor conformations
- Consensus scoring across independent scoring functions
- Pose clustering with interaction fingerprints
- Batch docking across full libraries, queued and monitored
In-Silico Suite · Docking resultsRanked poses, interaction diagram, consensus score breakdown. 04Molecular dynamics & binding free energy
A good score achieved in a pose that dissociates in two nanoseconds should not outrank a pose that holds. Dynamics is how you tell the difference.
- Explicit-solvent atomistic simulation with standard force fields
- Trajectory analysis: RMSD, RMSF, radius of gyration, hydrogen-bond occupancy, contact maps
- Binding free-energy calculation for series ranking
- Stability assessment across the trajectory, not just at the docked pose
- Full trajectory data available for your own reanalysis
In-Silico Suite · MD analysisRMSD trace, interaction occupancy heatmap, free-energy ranking across four series. 05Virtual screening
A staged cascade, cheapest filter first, so the expensive compute is only ever spent on compounds that have already survived something.
- Structure-based and ligand-based screening in a staged cascade
- Pharmacophore-driven pre-filtering
- Property and liability filters applied before compute is spent
- Diversity selection on the final hit list
- Sourcing availability surfaced with each hit
In-Silico Suite · Screening cascadeFunnel view showing compounds surviving each stage. 06ADMET & toxicity prediction
Absorption, distribution, metabolism, excretion and toxicity endpoints predicted as a profile rather than a single number.
- Physicochemical and pharmacokinetic property prediction
- Metabolic stability and clearance estimation
- Cardiotoxicity, hepatotoxicity, mutagenicity and structural-alert screening
- Confidence intervals and applicability-domain flags on every prediction
- Side-by-side series comparison
In-Silico Suite · ADMET profileEndpoint panel with confidence bands and a series comparison table. 07QSAR, property prediction & developability
Models built on your assay data, split properly, and honest about the chemistry they have never seen.
- Model building on your own assay data, with proper validation splits
- Applicability-domain assessment so you know when a model should not be trusted
- Drug-likeness, lead-likeness and rule-based filters
- Synthetic accessibility scoring with route feasibility signals
- Developability assessment combining potency, properties and liabilities
In-Silico Suite · QSAR validationModel validation plot with the held-out split. In-Silico Suite · DevelopabilityThe developability scorecard for a series. 08AI molecule generation & lead optimisation
Generation conditioned on a target, a scaffold or a property profile - and optimisation that shows you the trade-off instead of collapsing it into one score.
- De novo generation conditioned on a target, a scaffold or a property profile
- Scaffold hopping and bioisosteric replacement
- Multi-objective optimisation across potency, selectivity, ADMET and synthetic accessibility simultaneously
- Pareto-front exploration - see the trade-offs instead of collapsing them into one score
- Every generated design carries its predicted profile and a synthesisability assessment
In-Silico Suite · Generative designGenerated series with predicted profiles. In-Silico Suite · Pareto frontThe multi-objective trade-off surface. 09Pharmacophore, quantum chemistry & protein engineering
The specialist end of the workbench, sitting on the same project record as everything else.
- Pharmacophore model building from ligand sets or from a complex structure
- Quantum-chemical calculation for electronic properties and reaction energetics
- Protein engineering: mutation effect prediction, stability and affinity impact
- Structure-based design recommendations for the next chemistry cycle
In-Silico Suite · Pharmacophore modelModel overlaid on an aligned ligand set. 10Visualisation, workflows, copilot and reporting
Chain the modules into a protocol, run it across a library, and get out a document a non-specialist can read.
- Interactive 3D viewer for proteins, ligands, complexes, surfaces and trajectories
- Workflow automation - chain modules into a repeatable protocol and run it across a library
- AI copilot for method selection, parameter guidance and result interpretation
- Report generation: scientific reports with figures and methods, executive summaries for non-specialists
- Export to PDF, CSV, SDF and standard structure formats
In-Silico Suite · 3D viewerProtein–ligand complex with surface and trajectory playback. In-Silico Suite · Workflow builderA saved multi-stage protocol.
Virtual screening
Spend the compute last, not first
Every stage of the cascade is more expensive per compound than the one before it, so every stage is ordered to hand the next one less work. Hover or focus a stage to read what it does and what it costs.
01Library
2,000,000 compounds
The screening set as ingested: standardised, deduplicated, fingerprinted.
LibraryThe screening set as ingested: standardised, deduplicated, fingerprinted.02Property & liability filters
840,000 compounds1× cost/compound
Physicochemical range, drug-likeness, PAINS and structural alerts. Cheap arithmetic on descriptors already computed at ingest.
Property & liability filtersPhysicochemical range, drug-likeness, PAINS and structural alerts. Cheap arithmetic on descriptors already computed at ingest.03Pharmacophore pre-filter
220,000 compounds4× cost/compound
The feature geometry the target requires, applied before anything is docked.
Pharmacophore pre-filterThe feature geometry the target requires, applied before anything is docked.04Ligand-based similarity
48,000 compounds12× cost/compound
Fingerprint and shape similarity to known actives, ranked rather than thresholded.
Ligand-based similarityFingerprint and shape similarity to known actives, ranked rather than thresholded.05Structure-based docking
6,000 compounds900× cost/compound
Flexible-ligand docking into the prepared site, with pose clustering on interaction fingerprints.
Structure-based dockingFlexible-ligand docking into the prepared site, with pose clustering on interaction fingerprints.06Consensus rescoring
800 compounds3,200× cost/compound
Independent scoring functions run over surviving poses; agreement between them is the signal.
Consensus rescoringIndependent scoring functions run over surviving poses; agreement between them is the signal.07Diversity selection
240 compounds3,400× cost/compound
Scaffold-diverse selection with sourcing availability attached, so the list is orderable as well as interesting.
Diversity selectionScaffold-diverse selection with sourcing availability attached, so the list is orderable as well as interesting.
Schematic of the cascade shape, with illustrative counts. The stage order, not the numbers, is the point: docking a two-million-compound library directly is a way to spend a quarter of your compute budget on molecules a descriptor filter would have removed in seconds.
Lead optimisation
Move the weights and watch the ranking disagree with you
Multi-objective optimisation returns a front, not a winner. Re-weight the three objectives and the ranked list re-sorts underneath you - but the compounds on the front stay on the front, because no weighting makes a dominated compound the right answer.
Objective weights
Potency against ADMET
The ringed compounds are the front - nothing in the set beats them on both axes. Dot size is synthesisability.
Ranked at the current weights
12 candidates
- 01PGX-063Series Bfront0.75
- 02PGX-084Series Bfront0.73
- 03PGX-078Series Bfront0.73
- 04PGX-052Series Afront0.72
- 05PGX-091Series Cfront0.72
- 06PGX-041Series Afront0.69
- 07PGX-117Series C0.68
- 08PGX-103Series A0.61
- 09PGX-124Series C0.58
- 10PGX-136Series D0.53
- 11PGX-142Series D0.52
- 12PGX-155Series D0.45
Illustrative candidates. In a live project the axes are your predicted endpoints, the points are your designs, and every one of them opens into the model outputs and applicability-domain flags behind it.
Data and AI
What powers the predictions
| Domain | Sources |
|---|---|
| Compounds & bioactivity | ChEMBL, PubChem |
| Proteins & sequences | UniProt |
| Experimental structures | Protein Data Bank |
| Predicted structures | AlphaFold Protein Structure Database |
| Cheminformatics toolchain | RDKit-based standardisation, descriptors and fingerprints |
Every prediction reports a confidence estimate and an applicability-domain flag. A model that is being asked about chemistry it has never seen says so.
Methods and models
Physics-based simulation
Established docking engines, explicit-solvent molecular dynamics and free-energy methods - not surrogates for them.
Graph neural networks
Molecular graphs for property and activity prediction.
Chemical language models
De novo generation and scaffold transformation.
Gradient-boosted and ensemble models
ADMET endpoints, trained on curated public bioactivity data and extendable with your own.
Multi-objective optimisers
Pareto-front exploration across competing design objectives.
Retrieval-grounded language models
The copilot and report generation, cited to method records and result artefacts.
How a campaign runs
Seven stages, one parameter record
Select a stage to see what happens in it. Each one hands the next its output in place - there is no export step in the middle of this sequence.
Stage 1 of 7
Import — Compound library and target structure, standardised automatically.
Chemistry comes in as SDF, SMILES, MOL2, CSV or a direct database pull; the target comes from the PDB, from AlphaFold or from your own file. Standardisation, descriptor calculation and duplicate detection all run on ingest, so the project starts from one canonical form of every molecule.
What it hands on
- Standardised library
- Prepared target record
- Descriptors & fingerprints
Stage 2 of 7
Prepare — Protein preparation, binding-site definition, library filtering.
Missing residues repaired, protonation assigned, structure minimised, pockets detected and ranked. On the ligand side, the property and liability filters that will open the cascade are chosen and recorded here rather than improvised later.
What it hands on
- Prepared structure
- Binding site
- Filter set
Stage 3 of 7
Screen — Staged cascade with consensus rescoring and diversity selection.
The cascade runs cheapest-first: property and liability filters, pharmacophore pre-filter, ligand-based similarity, then docking on what survives. Consensus rescoring across independent scoring functions ranks the poses, and diversity selection keeps the final list from being one scaffold twelve times.
What it hands on
- Ranked hit list
- Pose clusters
- Sourcing availability
Stage 4 of 7
Simulate — Dynamics and free energy on the surviving series.
Explicit-solvent simulation on the poses that matter, with RMSD, RMSF, hydrogen-bond occupancy and contact maps reported across the trajectory. Binding free energy ranks the series; the trajectory itself is yours to reanalyse.
What it hands on
- Trajectories
- Stability profile
- Free-energy ranking
Stage 5 of 7
Profile — ADMET, liabilities, developability, synthetic accessibility.
Twenty-six-endpoint ADMET profiling with confidence intervals and applicability-domain flags, structural-alert screening, synthetic accessibility with route feasibility, and a developability assessment that combines potency, properties and liabilities into a scorecard rather than a number.
What it hands on
- ADMET profile
- Liability flags
- Developability scorecard
Stage 6 of 7
Optimise — Multi-objective design cycles against your target profile.
Matched molecular pairs, scaffold hopping, bioisosteric replacement and de novo generation, all scored against the same objective set. The optimiser returns a Pareto front, so the trade-off between potency, ADMET and synthesisability stays visible instead of being averaged away.
What it hands on
- Generated designs
- Pareto front
- Synthesisability assessment
Stage 7 of 7
Report — Scientific and executive outputs, with the full parameter record attached.
A scientific report with figures and methods, an executive summary for people who will not read it, and the complete parameter history behind both - every module version, every setting, every input file - so the campaign can be re-run and the result defended.
What it hands on
- Scientific report
- Executive summary
- Parameter record
Enterprise AI for Life Sciences
Ready to Advance Your Drug Discovery Pipeline?
Partner with Prognica Labs to leverage enterprise-grade AI, computational chemistry, and molecular simulation technologies that accelerate discovery, reduce development risk, and improve R&D productivity.
From biotech startups to global pharmaceutical organizations, we help research teams make faster, evidence-driven decisions across every stage of early drug discovery.