Large Databases of Small Molecules - Drug Development Tool and Public Resource

Division Of Basic Sciences - Nci

Investigators

Abstract

The Synthetically Accessible Virtual Inventory (SAVI) project is an initiative in computational chemistry and drug discovery. Its purpose is to accelerate the identification of new drug candidates by generating an immense virtual library of chemical compounds that are not only theoretically plausible but also reliably and affordably synthesizable in a laboratory. Traditional drug discovery often involves synthesizing compounds and then testing them or designing compounds virtually and then attempting their synthesis. The vastness of chemical space (the enormous number of possible molecules) means that only a tiny fraction can ever be explored experimentally. Furthermore, many theoretically designed molecules are difficult, expensive, or impossible to synthesize, leading to significant bottlenecks and wasted resources. SAVI tackles these issues by inverting the traditional approach. Instead of asking "How can I make this molecule I designed?", SAVI asks, "What can I easily and cheaply make?". It achieves this through a sophisticated, rule-based, forward-synthetic approach: Expert-System Chemical Reaction Rules (Transforms). SAVI is built upon a curated set of chemical reaction rules, referred to as "transforms." These rules are not merely generic reaction types; they are highly detailed, expert-system knowledge derived from decades of organic chemistry. Many of these rules originate from adaptations and extensions of the LHASA (Logic and Heuristics Applied to Synthetic Analysis) project, which was initially designed for retrosynthetic analysis (working backward from a target molecule to its precursors). For SAVI, these rules are cleverly adapted and applied in a forward-synthetic manner - taking starting materials and predicting the products of robust, high-yield reactions. These transforms incorporate rich chemical context annotation, including functional group reactivity data, expected yields, reliability, stereochemical outcomes, and necessary reaction conditions (e.g., solvent, temperature, catalysts). They also include "KILL" statements to filter out products that might lead to side reactions or be unstable, and "SCORE" statements to rate the expected yield and reaction complexity. The project uses a large inventory of commercially available chemical "building blocks" (starting materials). These are compounds that can be reliably purchased, often at reasonable prices, from chemical suppliers (e.g., Enamine, Sigma-Aldrich). Crucially, these building blocks are extensively annotated with practical information like pricing, availability, and chemical properties, further ensuring the "affordably synthesizable" aspect of the generated compounds. The CACTVS cheminformatics toolkit was originally used and is still used for virtual synthesis. It systematically applies the forward-synthetic transforms to the vast array of building blocks. This combinatorial approach allows the generation of billions of unique compounds from a relatively smaller set of starting materials and reactions. While the technology can handle multi-step reactions, initial and widely used SAVI datasets (like SAVI-2020) focused on single-step, two-reactant syntheses to maintain high confidence in synthesizability and manage the sheer scale. As compounds are generated, they are immediately characterized and filtered based on various "drug-likeness" properties. SAVI has generated virtual libraries containing billions of compounds (e.g., SAVI-2020 with over 1 billion, and more recently, SAVI-2025 encoding 5.5 billion). This enables the exploration of vast, previously unexplored regions of chemical space.