Virtual screening
Rank a compound library against your target online with docking and similarity tools, check the poses and ADMET predictions, and export a shortlist for testing.
More virtual screening workflows
Tools for virtual screening
Use these methods to prepare inputs, run the core analysis, inspect outputs, and validate the evidence described in this workflow.
ChEMBL Download
Retrieves compounds and activity records for reference sets and screening libraries.
PubChem Download
Downloads compound records and structures from PubChem for library assembly.
Molecular descriptors
Calculates physicochemical descriptors, Morgan fingerprints, and MACCS keys.
Open Babel
Converts, prepares, and standardizes molecular files across common formats.
SMILES to SDF
Generates SDF structures from SMILES for structure-based screening stages.
Veber's rule
Reviews molecular flexibility and polar surface area against Veber criteria.
PAINS filter
Flags PAINS substructures for compound-library review.
AutoDock-GPU
Runs GPU-accelerated docking for higher-throughput structure-based screens.
SMINA
Supports docking, minimization, and alternative scoring for comparative review.
GNINA
Generates ranked docking poses with CNN and Vina scoring outputs.
Admetica
Predicts ADMET properties for post-screening prioritization.
Structural alert screening
Screens shortlisted compounds against PAINS, Brenk, and NIH structural alert catalogs.
What is virtual screening?
Virtual screening is a computational method for ranking a compound library against a target, so that only the best-ranked compounds go on to purchase and laboratory testing. A library of a million compounds is far too large to assay, but a computer can score all of it in days. The shortlist that comes out might be a few hundred compounds, small enough to order and test.
The ranking method depends on what you start with. A reliable 3D structure of the target supports docking: each compound is placed in the binding pocket and scored for fit. A set of known active compounds supports similarity searching instead, which ranks the library by resemblance to those references and needs no target structure at all. Many campaigns chain the two, using a fast filter first and slower docking on what survives.
The result is a prioritized list, not a list of binders. A docking score is a model's estimate under one setup, and even well-run screens return many top-ranked compounds that are inactive in the assay. What a screen offers is enrichment: the active compounds are more likely to sit near the top than in a random pick. Every virtual screen ends in an experiment.
When to use virtual screening
- Narrow a large purchasable or in-house library to a set small enough to test.
- Look for new scaffolds against a target with a solved or well-predicted structure.
- Expand a hit series by finding compounds similar to known actives.
- Check one compound against a panel of targets for off-target or repurposing hypotheses.
Strengths and limitations
A virtual screen is cheap compared with screening the same library in an assay, and it can cover chemistry that no physical collection holds. Because every step is a recorded calculation, a hit or a miss can be traced back to the structure, settings and score that put the compound on the shortlist.
The ranking is only as good as its inputs. Protonation state, tautomers, stereochemistry, receptor conformation and pocket definition can each reorder a list, and scores from different methods are not on the same scale. A computational ranking does not confirm binding, activity, selectivity or safety.
Types of virtual screening
The main types differ in the evidence they start from. Structure-based virtual screening docks each compound into a prepared target pocket, so it needs a good receptor structure and a defined binding site. Ligand-based virtual screening ranks compounds against known actives and works when no usable structure exists. A docking score and a similarity value measure different things and should not be merged into one number.
Shape-based virtual screening looks for compounds whose 3D envelope matches a reference ligand in its bound conformation. Pharmacophore-based virtual screening adds the spatial arrangement of features such as hydrogen-bond donors and aromatic rings, which can retrieve a different scaffold than a fingerprint search.
High-throughput virtual screening is a scale strategy, not a separate scoring method: it reserves expensive calculations for a small subset after cheap filters have run. Inverse virtual screening turns the question around and tests one compound against many targets.
Which virtual screening tool to use
Each tool below runs online and quotes its credit cost before it starts. The task you are doing decides which one you need.
| What you want to do | Tool | What you get back |
|---|---|---|
| Build a library from known compounds | ChEMBL Download and PubChem Download | Compound records and structures |
| Remove problem compounds before screening | PAINS Filter | Flagged substructures |
| Prepare the receptor and find pockets | PDBFixer and fpocket | A repaired structure and ranked pockets |
| Dock a library into a pocket | GNINA, AutoDock Vina, AutoDock-GPU | Ranked poses with each engine's own score |
| Rank compounds against a protein sequence | SPRINT | A cosine similarity score per compound |
| Check whether a docked pose is physically sensible | PoseBusters and ProLIF | Geometry checks and residue-level interactions |
| Predict drug-like properties for the shortlist | ADMET-AI | Per-compound ADMET predictions |
GNINA, AutoDock Vina and AutoDock-GPU are different docking engines with different scoring functions. If you dock the same library with more than one, keep their scores in separate columns instead of averaging them.
Virtual screening services and software
Searches for virtual screening services usually mean one of two things: a company that runs the screen for you, or software you run yourself. ProteinIQ is the second. You upload a target and a library, choose the tools, and get every score, pose and log back to inspect.
If you want someone else to design the campaign, prepare the target, run the screen and interpret the ranking, that is a separate done-for-you project, and contact sales to scope it. Published prices for managed screening range widely with the deliverable, so the FAQ below gives examples rather than a single rate.
How to do virtual screening online
Start with a decision and a validation plan, not with the biggest library you can find.
- Define the question. Write down the target, the reference evidence you have, the size of the library and how many compounds you can afford to test.
- Set aside a validation subset. Include known actives and inactives or decoys if you have them, so you can check that the method ranks actives above the rest before you scale it.
- Standardize the inputs. Keep identifiers while you fix salts, stereochemistry, protonation and tautomers, and keep a list of the compounds you drop.
- Run the featured workflow. It takes a target sequence, a target PDB and a SMILES library, such as the FDA-approved drug library for a repurposing screen. SPRINT ranks the compounds by cosine similarity, and a separate docking branch runs the same library into the repaired structure. Check the docking pocket setting, because the default focused search needs a positioned reference ligand.
- Inspect rankings and failures. Review the docked poses with PoseBusters and ProLIF, then look at calculated properties, missing outputs and chemical diversity before you set a cutoff.
- Export and follow up. Keep the full run record, then test the shortlist with an orthogonal calculation and an assay suited to the biology.
How to interpret virtual screening results
Read each score in the context of the method that produced it. SPRINT's cosine similarity ranks protein and compound representations; GNINA's CNN score, predicted affinity and Vina score rank docked poses under a given receptor and pocket. A high value in one branch cannot be converted into the other branch's scale, or into a binding probability.
Look at controls, invalid records and missing outputs before choosing a cutoff. Keep several scaffolds on the shortlist, with the evidence behind each, so that an assay hit or miss can be traced back to its preparation and ranking.
How the featured workflow works
The featured workflow runs a learned target-aware ranking and structure-based docking as separate evidence branches. Use a method-specific workflow instead when the program lacks a suitable target structure or protein sequence.
- Provide target evidence. Submit a compatible protein sequence for SPRINT and a reviewed PDB structure for the structure-based branch.
- Run target-aware ranking. Rank the submitted SMILES library with SPRINT and retain its native cosine similarity alongside compound identifiers.
- Prepare and dock. Repair the receptor with PDBFixer and dock the library independently with GNINA.
- Review independent evidence. Check GNINA geometry with PoseBusters, calculate residue-level interactions with ProLIF, and review descriptors and ADMET predictions.
- Compare and shortlist. Compare the branches without treating their raw scores as interchangeable, then export a diverse shortlist for orthogonal review.
Inputs and outputs for the featured workflow
Check formats before running, then inspect and download the result from every workflow step.
Inputs
Target structure
PDBThe main template accepts a PDB receptor. If the source structure is CIF, convert it to PDB before starting the run.
Compound library
SMILESSubmit a SMILES library directly. Convert existing SDF with SDF to SMILES or MOL2 with MOL2 to SMILES before launching; use the ligand-preparation variant to generate or repair structure files from SMILES.
Binding-site context
Known pocket coordinates or a reference ligand can inform manual docking setup and review, but they are not required inputs on this template.
Outputs
Prepared target and pocket results
PDBJSONDownload the fixed receptor and preparation provenance, plus fpocket pocket files and metrics.
Property and rule tables
CSVJSONSMILESInspect per-compound Lipinski and lead-likeness results with the values and warnings returned by each step.
Docking poses and scores
SDFCSVJSONReview ranked docking poses with GNINA CNN score, CNN affinity (pKd), and Vina score (kcal/mol).
Pose and ADMET review
CSVJSONInspect downloadable PoseBusters validation tables and ADMET-AI prediction tables as separate evidence layers.
Run record
LOGFILESRetain settings, node-level results, logs, and downloadable files for review or handoff.
Frequently asked questions
Start with a small, representative set that includes compounds with known or expected behavior when possible. Review target and compound preparation, binding-site context, failure rates, pose quality, and ranking behavior before committing a larger library to the same setup.
Current public examples range from $500 for a limited university-core computational screen to $5,000 for a published 1.76-million-compound 3D shape-screening package, $15,000 for an integrated target-assessment project that includes virtual screening, and $20,000 for one service exploring up to 10 billion compounds.
These are unlike deliverables rather than a universal per-compound rate. Library preparation, method, target or reference complexity, compute settings, shortlist analysis, compound delivery, and experimental validation account for much of the difference.
For academic use, ProteinIQ’s Free plan gives 200 credits at sign-up and refills to 100 credits each month; paid plans start at $9 per month for Lite, with Plus at $29 per month and Pro at $99 per month, which adds a commercial license. The configured workflow is quoted in credits before it runs. Done-for-you projects are scoped separately.
For academic use, yes within the Free plan limits: 200 credits at sign-up, a 100-credit monthly refill and three jobs a day. Credits are quoted before each run, so a small library or a validation subset fits before you scale up on a paid plan.
Yes. Workflow templates are editable, so you can replace compatible preparation, docking, filtering, or prediction steps and adjust their settings. Keep engine-specific scores and assumptions visible when comparing alternative workflow versions.
They can support a consensus strategy, but raw scores from unrelated methods should not be treated as directly interchangeable. Define the combination rule in advance, preserve each method’s original score, and check whether the resulting consensus improves ranking on an appropriate validation set.
Use docking scores for prioritization within a defined setup, not as measured binding affinities or binding free energies. Compare engines in their own scoring context. Treat ADMET and toxicity predictions as triage signals with model-specific uncertainty, not proof of safety or clinical behavior.
Review the underlying structures, poses, model applicability, and failure checks before selecting compounds for follow-up. Shortlisted compounds still require orthogonal computational review and experimental assays appropriate to the target, mechanism, and development question.
AI answers from our docs. No sign-in needed.
Run your first analysis today
Start with 200 free credits and no installs. Upgrade when your project needs more compute.