From predictions to the wet lab
Translating in-silico predictions into wet lab experiments, from hit prioritization through lead optimization and CRO engagement.
The platform produces ranked compound lists, predicted binding poses, ADMET profiles, and synthesis routes. This page outlines how to convert those computational outputs into an experimental validation plan -- from cherry-picking hits for synthesis through dose-response screening, lead optimization, and eventually IND-enabling studies.
This guide describes a general small-molecule workflow. Biologics (antibodies, peptides, oligonucleotides) follow a parallel path with different assay types but the same prioritization logic.
From analysis to wetlab execution
The computational-to-experimental gap
In-silico predictions are hypotheses, not facts. Every docking score, ADMET prediction, and de novo design must be validated experimentally before it can inform a drug program. The goal of this roadmap is to minimize wasted synthesis and assay cycles by applying rigorous triage criteria to computational outputs before committing wet lab resources.
Workflow overview
Planning experimental validation
Define success criteria up front
Before synthesizing or purchasing any compounds, agree on quantitative go/no-go thresholds:
| Parameter | Typical threshold | Notes |
|---|---|---|
| Biochemical IC50 | < 1 uM | Depends on target class |
| Cell-based EC50 | < 10 uM | Counter-screen for cytotoxicity |
| Selectivity ratio | > 30x over closest homolog | Critical for kinases, GPCRs |
| Aqueous solubility | > 10 ug/mL | Minimum for in vitro assays |
| Microsomal stability | t1/2 > 30 min (human) | Enables oral dosing |
| hERG IC50 | > 10 uM | Safety margin for cardiac liability |
Experimental design considerations
- Positive and negative controls: Include known active and inactive compounds in every assay plate. The platform's compound passport reports literature-sourced reference values that can guide control selection.
- Replicate strategy: Run a minimum of 3 biological replicates for dose-response curves. Technical triplicates are standard for single- point screens.
- Blinding: Where possible, blind the experimentalist to computational predictions to avoid confirmation bias in data interpretation.
Prioritizing computational hits for synthesis
Triage criteria
Apply filters to the platform export in order of cost:
- Structural alerts -- Remove compounds flagged as PAINS, reactive metabolites, or DNA-intercalating. The platform's ADMET tools flag these automatically.
- Predicted ADMET -- Exclude compounds with predicted poor solubility (< 1 ug/mL), high CYP inhibition risk, or hERG liability. Use the platform's ADMET-AI outputs.
- Synthetic accessibility -- Prioritize compounds with a synthetic accessibility (SA) score below about 4 on the 1-10 Ertl-Schuffenhauer scale, and with a solved AiZynthFinder route. Higher SA scores and unsolved routes point to multi-step or low-yielding chemistry that raises cost and delays timelines.
- Chemical diversity -- Cluster remaining hits by Tanimoto similarity (ECFP4, cutoff 0.4) and select 1-2 representatives per cluster. This maximizes SAR information per dollar spent.
- Docking pose quality -- For structure-based campaigns, inspect the top docking poses in the platform's 3D viewer. Discard compounds where the predicted binding mode places key pharmacophoric groups in solvent rather than in the binding pocket.
Procurement vs. custom synthesis
| Source | Turnaround | Cost per compound | When to use |
|---|---|---|---|
| Commercial vendors (e.g. Enamine, Mcule) | 1-3 weeks | $50-$300 | Hit validation, SAR exploration |
| Make-on-demand (Enamine REAL) | 3-6 weeks | $300-$1,000 | Near-analogs not in stock |
| Custom CRO synthesis | 6-12 weeks | $2,000-$10,000+ | Lead optimization, novel scaffolds |
Start with commercially available analogs to validate the scaffold before investing in custom synthesis. The platform's AiZynthFinder routes flag whether each building block is in its stock set (a ZINC in-stock snapshot); confirm current price and availability with the vendor before ordering.
Primary screening
Binding assays
- SPR (Surface Plasmon Resonance): Measures real-time binding kinetics (kon, koff, KD). Best for confirming predicted binding and ranking hits by affinity.
- TSA (Thermal Shift Assay): A low-cost, high-throughput screen. Compounds that stabilize the target protein shift its melting temperature (delta-Tm > 1C is a positive hit).
- FP (Fluorescence Polarization): Competitive binding assay useful when a fluorescent probe ligand is available.
Activity assays
- Enzymatic assays: For enzyme targets, measure substrate turnover in the presence of compound. IC50 determination requires an 8-point dilution series.
- Cell-based reporters: For intracellular targets, use reporter gene assays (luciferase, GFP) to measure pathway modulation.
Dose-response and selectivity screening
Dose-response curves
Confirmed hits from primary screening advance to full dose-response characterization:
- Concentration range: Test 8-10 half-log dilutions spanning 4-5 orders of magnitude (e.g., 100 uM to 1 nM).
- Curve fitting: Fit data to a four-parameter logistic (4PL) model to derive IC50/EC50, Hill slope, and top/bottom plateaus.
- Quality metrics: Require R^2 > 0.95 and Hill slope between 0.5 and 2.0. Steep Hill slopes (> 2.0) may indicate aggregation or assay artifacts.
Selectivity profiling
- Counter-screens: Test against 3-5 closely related family members (e.g., kinase panel for kinase inhibitors).
- Broader panels: Services like Eurofins SafetyScreen44 or DiscoverX scanMAX profile compounds against hundreds of targets to identify off-target liabilities early.
- Platform integration: Compare experimental selectivity data against the platform's predicted binding poses across homologous structures to build a computational SAR model.
Lead optimization cycle
The design-make-test-analyze loop
Lead optimization is iterative. Each cycle should take 2-4 weeks:
- Design -- Use the platform's de novo design (REINVENT4, GenMol) and docking tools to propose analogs that improve potency, selectivity, or ADMET properties.
- Make -- Synthesize 5-15 analogs per cycle. Prioritize changes to one region of the molecule at a time to build clean SAR.
- Test -- Run the standard assay cascade: binding, cell activity, solubility, microsomal stability, and permeability (PAMPA or Caco-2).
- Analyze -- Feed experimental results back into the platform. Update compound passports with measured data and retrain QSAR models if available.
Key optimization targets
| Property | Strategy | Platform tools |
|---|---|---|
| Potency | Grow into sub-pockets | DiffDock |
| Selectivity | Exploit residue differences | Sequence alignment, docking |
| Metabolic stability | Block soft spots (benzylic, phenol) | ADMET-AI |
| Solubility | Add polar handles, reduce planarity | RDKit descriptors, GraphSol |
| Permeability | Reduce MW, HBD count, polar surface area | ADMET predictions |
Avoid optimizing a single property in isolation. Improving potency by adding lipophilic groups often worsens solubility and metabolic stability. Track a multi-parameter optimization (MPO) score across every cycle.
Working with CROs
Selecting a contract research organization
| CRO type | Services | Example providers |
|---|---|---|
| Chemistry | Custom synthesis, medicinal chemistry | WuXi, Pharmaron, Evotec |
| Biology | Assay development, screening | Charles River, Eurofins |
| DMPK | In vitro ADME, PK studies | Absorption Systems, Sekisui XenoTech |
| In vivo | Efficacy, toxicology | Covance, Inotiv |
Best practices for CRO engagement
- Provide clear specifications: Include the platform's synthesis route (AiZynthFinder output), target purity (> 95%), and required quantity (typically 10-50 mg for in vitro, 500 mg+ for in vivo).
- Request analytical data: Require LCMS, 1H-NMR, and HPLC purity traces for every synthesized compound.
- Establish compound tracking: Assign platform compound IDs to every CRO-synthesized batch. Register batches in your ELN with links back to the platform project.
- Milestone-based contracts: Structure CRO agreements around deliverables (number of compounds, assay data packages) rather than time-based billing.
Regulatory considerations
Data integrity and traceability
Regulatory agencies (FDA, EMA) expect full traceability from computational prediction through experimental validation:
- Audit trail: The platform's job history and
_metadata.jsonexports provide timestamped records of every computation. Archive these alongside experimental data in your regulatory submission dossier. - GLP compliance: For IND-enabling toxicology studies, ensure the CRO operates under Good Laboratory Practice. Computational predictions do not require GLP but should be documented in the study report.
- Electronic records: If your organization follows 21 CFR Part 11, ensure that platform exports stored in your ELN meet electronic signature and audit trail requirements.
IND-enabling studies
Before filing an Investigational New Drug (IND) application, the following experimental packages are typically required:
- In vitro pharmacology -- Target engagement, potency, selectivity.
- In vitro ADME -- Metabolic stability, CYP inhibition/induction, plasma protein binding, permeability.
- In vivo PK -- Single-dose pharmacokinetics in at least one rodent and one non-rodent species.
- GLP toxicology -- 14-day or 28-day repeat-dose toxicity in two species with toxicokinetics.
- Genotoxicity -- Ames test and in vitro micronucleus assay.
- CMC -- Chemistry, Manufacturing, and Controls for GMP synthesis.
The platform's compound passport aggregates predicted ADMET, safety signals from FAERS, and synthesis feasibility data in a single report. This is a useful starting point for IND planning but does not substitute for experimental GLP studies.
Feedback loop: experimental data back into the platform
The most valuable part of the wet lab transition is feeding measured data back into computational models:
- Update compound records: Add experimental IC50, solubility, and PK parameters to platform compound entities. This enriches future passport reports and improves team-wide visibility.
- Retrain models: If you maintain custom QSAR models, incorporate new experimental data points to improve prediction accuracy for your chemical series.
- Refine docking models: Compare experimental binding data (SPR KD values) against predicted docking scores. Recalibrate scoring functions or select alternative protein conformations if correlation is poor.
- Archive and share: Publish finalized datasets (computational + experimental) to platform projects so collaborators can build on validated results rather than re-running predictions.
Summary timeline
| Phase | Duration | Key deliverables |
|---|---|---|
| Hit triage | 1-2 weeks | Prioritized compound list (20-50 compounds) |
| Procurement / synthesis | 2-6 weeks | Compounds in hand, QC data |
| Primary screen | 1-2 weeks | Confirmed hits (5-15% hit rate typical) |
| Dose-response | 2-3 weeks | IC50/EC50 values, selectivity panel |
| Lead optimization (per cycle) | 2-4 weeks | 5-15 analogs tested per round |
| In vivo PK | 4-6 weeks | Plasma exposure, half-life, bioavailability |
| IND-enabling | 6-12 months | GLP tox, CMC, IND filing |
Was this page helpful?