What is virtual screening?

Virtual screening is the computational counterpart of high-throughput screening. Instead of testing every compound in a library, software scores the library and the team tests only the top-ranked molecules. A good virtual screen finds more active compounds per experiment than random selection.

Two families of methods

Ligand-based

If you already know at least one active molecule, you can search for compounds that resemble it. The usual tools are molecular fingerprints compared with Tanimoto similarity, substructure search, pharmacophore models and 3D shape matching. Ligand-based methods are fast — billions of comparisons are feasible — but tend to find molecules similar to what you already have.

Structure-based

If a 3D structure of the target is available, from crystallography, cryo-EM or structure prediction, each compound can be docked into the binding site and ranked by a scoring function. Docking can find chemically novel actives but is slower and its scores are noisy.

Measuring success: enrichment

Virtual screens are judged by enrichment: how many more actives appear in the top of the ranked list than chance would give. An enrichment factor of 10 at 1% means the top 1% of the list contains ten times as many actives as a random 1% would. Retrospective benchmarks use known actives mixed with “decoys”; the real test is always a prospective experiment.

A practical workflow

1. Choose the library. In-stock compounds can be delivered in weeks; virtual (make-on-demand) libraries are far larger but must be synthesised.
2. Filter. Remove compounds with reactive or assay-interfering groups and those outside the property range you need.
3. Rank cheaply first. Use similarity or substructure search to reduce millions of molecules to thousands.
4. Rank expensively second. Dock the shortlist and inspect the best poses by eye.
5. Pick for diversity. Choose a varied set rather than the top 50 near-duplicates, then order and test.