Accelerating Discovery: Thompson Sampling for Virtual Screening Now Available on Vecura

What is Thompson Sampling for Virtual Screening?

Thompson Sampling for Virtual Screening is an active-learning algorithm designed to navigate massive combinatorial chemistry libraries, which can contain billions or even trillions of potential molecules. Instead of attempting to enumerate and evaluate every possible product, the model treats each chemical reagent as a Bayesian arm, iteratively sampling and scoring combinations based on their predicted performance. This intelligent exploration allows users to identify top-scoring candidates efficiently while only analyzing a tiny fraction of the total chemical space.

It helps users discover high-potential drug candidates without the massive computational cost of full library enumeration. It is especially useful for medicinal chemists and computational researchers working with large-scale "synthesis-on-demand" databases who need to prioritize molecules for real-world synthesis.

What can users do with Thompson Sampling on Vecura?

With Thompson Sampling for Virtual Screening on Vecura, users can:

• Define chemical reactions using standard SMARTS notation to explore combinatorial libraries.
• Optimize for specific molecular properties, such as similarity to a target scaffold or molecular weight.
• Efficiently search ultra-large reagent pools with minimal computational overhead.
• Recover top-tier molecules while inspecting less than 0.1% of the total available library space.

What the output means

The output provides a prioritized list of the top-scoring unique product molecules, complete with their computed scores and SMILES strings. Additionally, it provides metrics on the search efficiency, including the total number of evaluations performed and the percentage of the full library space covered.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

Navigating "chemical space" is a primary bottleneck in modern drug discovery. When dealing with billions of possible molecules, brute-force enumeration becomes computationally prohibitive and scientifically impractical. By leveraging Bayesian optimization, this tool bridges the gap between massive theoretical libraries and feasible experimental design.

By concentrating evaluation resources on the most promising reagents, researchers can drastically accelerate the discovery of novel compounds. This approach transforms the virtual screening workflow from a search for "needles in a haystack" into a targeted exploration, significantly increasing the probability of finding viable lead compounds early in the drug discovery cycle.

• Developed by: Pat Walters et al.
• Source: GitHub / ACS Publications
• Reference: Thompson Sampling, an Efficient Method for Searching Ultralarge Synthesis-on-Demand Databases (JCIM 2024)