No Reagents for the In Silico Phase
Confidence-Scored Results
Publication-Ready Figures
Validation Planned With Partners
Response Within One Business Day
Service Overview
ICDMO's AI Multimodal Protein–Compound Screening Service employs machine learning and deep learning to analyze protein–small molecule interactions, leveraging three-dimensional protein structures to predict binding poses and affinities with superior speed and precision. The workflow combines batch structure-based docking, machine-learned rescoring and multimodal protein representation (three-dimensional structure, sequence and physicochemical properties) to produce a ranked, pose-inspected shortlist. Applicable to drug mechanism investigation, enzyme substrate recognition, signal transduction analysis, and lead optimization.
Technical Advantages
✓
Tiered High-Throughput Screening
Batch structure-based docking with machine-learned rescoring, run as a fast pre-filter followed by finer scoring on the survivors. Library size and timeline are confirmed against available compute during scoping.
✓
Fewer Compounds to Test
Computational triage narrows a large set to a defensible shortlist before expensive experimental screening begins. It reduces what you test; it does not replace testing.
✓
Multimodal AI Integration
Combines protein 3D structure, sequence features, and small molecule physicochemical properties for comprehensive interaction analysis.
✓
Transparent Score-Based Ranking
Results are ranked by docking score, which orders compounds relative to one another rather than predicting affinity. Top models are visualised with 3D views and 2D interaction diagrams.
✓
Flexible Input Formats
Accepts protein 3D structures or amino acid sequences; small molecules in 2D or 3D formats (mol2, sdf, pdb, SMILES).
✓
Seamless Downstream Validation
Top hits from virtual screening validated experimentally via SPR, MST, BLI, ITC, and cell-based assays.
Key Applications
Protein–Small Molecule Compound Docking
▸Drug mechanism investigation: understand how lead compounds engage target binding pockets
▸Enzyme–substrate recognition: identify substrate specificity determinants and inhibitor binding modes
▸Signal transduction: map allosteric sites and modulator binding to signaling proteins
▸GPCR and ion channel compound screening
Protein–Protein Interaction (PPI) Modulation
▸Identify small molecules that disrupt or stabilize protein–protein interfaces
▸Screen compound libraries against PPI hot spots for inhibitor discovery
▸AI-based analysis of PPI networks to discover novel drug targets
Virtual Screening & Hit Identification
▸Screen FDA-approved compound libraries for drug repurposing opportunities
▸Screen your own in-house compound collections against novel targets
▸Fragment-based screening for lead generation
▸Natural product library screening
Sample Submission Requirements
1Protein: 3D structure (PDB file) preferred; amino acid sequence accepted (structure will be modeled)
2Small molecules: 2D or 3D structural files (mol2, sdf, pdb, or SMILES string)
3Optional: Known binding pocket coordinates (box center + dimensions) for focused docking
4For large compound libraries (>10K): provide as SDF or SMILES database file
Docking Pipeline (Case Study: CYP51 + FDA Compounds)
1Step 1 — Structure preparation: the receptor is protonated, tautomer and flip states are assigned and structural waters are handled; ligands are ionised at assay pH with stereoisomers and conformers enumerated
2Step 2 — Define docking box: Confirm box size and center coordinates around active site (e.g., CYP51 heme binding cavity)
3Step 3 — Batch docking: every compound is docked into the defined site, producing a score and a binding pose for each
4Step 4 — Ranking and selection: Sort all compounds by binding affinity (kcal/mol); select top 5 models for detailed analysis
5Step 5 — Generate visualizations: 3D distant view, close-up interaction view, and 2D interaction diagram for each top candidate
What We Need From You
1Target structure (experimental or predicted) and the binding site to be screened
2The compound library to screen, or a request that we select and prepare a commercial or in-house set
3Known actives and inactives, which allow the protocol to be validated by retrospective enrichment before the real screen
4Selection constraints: purchasability, budget per compound, supplier preferences, chemistry to avoid
5Your assay format and throughput, which determine how large a shortlist is actually useful
How the Analysis Is Performed
1Receptor preparation: protonation and tautomer assignment, structural water handling, site definition
2Ligand preparation: ionisation at assay pH, tautomers, stereoisomer enumeration and conformer generation
3Property, drug-likeness, PAINS and aggregator pre-filtering with thresholds agreed in advance
4Staged docking from fast scoring through to precision scoring on the surviving fraction
5Multimodal rescoring that combines physics-based scores with learned models rather than relying on either alone
6Scaffold clustering, visual pose inspection and optional MM/GBSA rescoring of the final set
What You Receive
1Ranked hit list with scores, predicted properties and supplier catalogue identifiers where applicable
2Docked poses in SDF for the full shortlist, with a viewer session for inspection
3Per-compound interaction analysis: hydrogen bonds, hydrophobic contacts and any unfavourable features
4Scaffold clustering and diversity summary showing which chemotypes the shortlist covers
5Screening report with the protocol, filter thresholds, enrichment validation and realistic hit-rate expectations
Interpretation and Limitations
Docking and learned scores rank plausibility, not potency. Compounds separated by small score differences should be treated as indistinguishable, which is why diversity selection matters more than the top of the list. Hit rates vary widely between targets and cannot be promised in advance; where known actives exist we report the enrichment achieved on them so the expectation is grounded in your own system.
Questions We Are Asked Before Scoping
How large a library can you screen?
▸From curated sets of ten thousand through to make-on-demand spaces of a billion or more. Beyond roughly ten million compounds we switch to machine-learning triage with conventional re-docking of the survivors.
Do we need a crystal structure?
▸No, a predicted or modelled structure is often sufficient, but the quality of the binding site region governs the outcome. We assess site quality separately from overall model quality and say whether it will carry a screen.
What hit rate should we expect?
▸Published campaigns commonly report a few percent for well-prepared systems, with a wide spread. We do not promise a rate; we report the retrospective enrichment achieved on your known actives when you have them.
Technical Workflow
1
Scientific Consultation
We discuss the objective, the decision the result has to support and the data you already hold. Nothing is quoted before that is clear.
2
Feasibility Review
We check that your inputs, target class and available evidence can support the approach, and say plainly where they cannot.
3
Technical Proposal
Scope, methods, deliverables, timeline and price are confirmed in writing, along with the limitations we expect to report.
4
Computation
The library is prepared and filtered, the docking cascade runs from fast to precise scoring, and survivors are rescored, clustered and inspected pose by pose.
5
Report and Follow-Up
You receive the ranked shortlist with poses, interaction analysis, diversity summary and purchasing information, plus the protocol and its validation.
Service Timeline & Deliverables
Is This the Right Service for You?
Choose this when
A compound set needs ranking against a target structure before assay work
Both physics-based and learned scoring should inform the same shortlist
A purchasable, pose-inspected list is needed quickly
Consider another service when
Billion-scale make-on-demand space is the goal
AI-Powered Ultra-Large Library Screening One compound set needs binding-mode analysis rather than ranking
Molecular Docking What Goes Wrong, and How We Handle It
These are the failure modes that produce confident-looking results which do not survive the next experiment. Each one is designed against rather than discovered afterwards.
Machine-learned scores trusted without physics-based confirmation
Learned scores triage; everything on the shortlist is re-docked conventionally and inspected pose by pose.
A shortlist of one chemotype
Selections are clustered so several independent scaffolds reach the assay.
Screening a site that cannot support a ligand
Site quality is assessed before compute is committed, and we say when a screen is not worth running.
A Worked Example
Repurposing screen against an enzyme active site
The situation
A team wanted approved and clinical-stage compounds ranked against a newly characterised enzyme.
What the work involved
The receptor was prepared with explicit protonation decisions, the compound set was prepared with ionisation and tautomers enumerated, staged docking was run with learned rescoring, and every top pose was inspected.
What came back
Thirty-one compounds with plausible poses and documented interactions, clustered into seven chemotypes, with those relying on an unresolved side chain flagged as lower confidence.
Illustrative of the shape this work takes. Details are generalised, no client or target is identified, and scope and outcome vary with the system in question.
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 22% confirmed hit rate from docking 138 million compounds
In a landmark 2019 study, ~138 million make-on-demand compounds were docked against the dopamine D4 receptor. Of 549 compounds selected and tested, 122 were confirmed active — a 22% hit rate — including ligands in the sub-nanomolar range. The same work screened ~99 million compounds against AmpC β-lactamase, where 5 of 44 tested compounds were confirmed (11%), one of which was among the most potent non-covalent AmpC inhibitors reported at the time.
Lyu et al., Nature, 2019 ↗Hit rate rises with rank, which is why library scale matters
The same body of work showed that the probability of a compound being confirmed active increases as its docking rank improves, and that larger libraries push better-scoring molecules into the selected set. This is the practical argument for screening make-on-demand space rather than a physical catalogue — and also the reason scoring accuracy, not library size alone, remains the limiting factor.
Lyu et al., Nature, 2019 ↗Service Process
1
Online Consultation
2
Solution Matching
3
Service Contract
4
AI Computation
5
Project Report
Note: All services are for research use only and not intended for diagnostic or therapeutic purposes.
Get a Custom Quote
Our scientific team responds within 24 hours with a detailed technical proposal and pricing tailored to your research goals.
Contact Us Online ConsultationStandard Deliverables
✓Detailed analysis report (PDF)
✓Raw data files & processed outputs
✓High-resolution publication figures
✓Interaction scoring tables (Excel)
✓Project summary presentation
Related AI Services
Molecular Dynamics SimulationProtein Interaction & Binding Site PredictionDigital Library-to-Library ScreeningDigital Protein/Nucleic Acid Interaction ModelTarget-based Drug DiscoveryDrug Screening & ProfilingWhy ICDMO AI?
Computational CDMO Focus
Reproducible, Versioned Pipelines
Confidence & Limitations Reported
Response Within One Business Day