Retrosynthesis Made Simple: AiZynthFinder is Now Available on Vecura

What is AiZynthFinder?

AiZynthFinder is an open-source, automated retrosynthetic planning framework developed by the MolecularAI team at AstraZeneca. By combining Monte Carlo tree search (MCTS) with deep neural networks trained on millions of chemical reactions, it recursively deconstructs target molecules in reverse to identify viable precursor compounds. This framework systematically maps out chemical pathways until every starting material is verified against commercial inventory databases.

It helps users design and evaluate multi-step synthetic routes to novel organic compounds without relying solely on manual expertise. It is especially useful for early-stage drug discovery, enabling research teams to screen large virtual libraries for synthesizability and find cost-effective pathways for lead candidates.

What can users do with AiZynthFinder on Vecura?

With AiZynthFinder on Vecura, users can:

• Plan Multi-Step Retrosynthesis: Run deep tree searches to automatically break down a target molecule (SMILES) into starting materials that are commercially available in catalogs like ZINC.

• Customize Search & Filter Constraints: Control search depth, iteration limits, and algorithm backends (MCTS, Retro*, or DFPN), and utilize filter policies to discard synthetically implausible reaction paths.

• Combine Expansion Policies: Leverage standard USPTO models for general transformations alongside specialized policies like RingBreaker to capture complex ring-forming steps.

• Compare and Score Synthetic Routes: Rank potential synthetic pathways using multiple built-in metrics, including overall route cost, reaction step counts, and precursor stock availability.

• Perform Single-Step Disconnection Explorations: Quickly check the immediate precursor space of a scaffold to view reaction templates and feasibility scores without committing to a full multi-step search.

What the output means

The output provides an interactive, ranked list of retrosynthetic routes represented as step-by-step reaction trees, alongside comprehensive metrics including target synthesizability status (solved/unsolved), reaction template metadata, and precursor stock classifications. A companion route scoring table evaluates each pathway across selected criteria—such as state score, precursor cost, and step counts—allowing users to easily weigh the trade-offs of different synthetic approaches.

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

Why this matters

In modern drug discovery, the bottleneck has shifted from generating promising molecular designs to physically synthesizing them in the laboratory. Traditional retrosynthetic planning is heavily dependent on the subjective intuition of expert chemists and tedious, manual literature searches. Consequently, evaluating the feasibility of hundreds of candidate molecules is slow and resource-intensive, often leading to costly failures during wet-lab validation.

By automating retrosynthetic route exploration, AiZynthFinder shifts chemical synthesis planning from a manual art into a scalable, data-driven science. Integrating this technology directly into the Vecura workflow enables chemists to prioritize molecules that are genuinely accessible, optimize precursor purchasing, and significantly compress the time from design to physical assay.

• Developed by: MolecularAI, AstraZeneca

• Source: Official GitHub Repository | Documentation

• Reference: Genheden, S., Thakkar, A., Chadimová, V. et al. AiZynthFinder: a robust, open-source framework for retrosynthetic planning. ChemRxiv (2020). DOI: 10.26434/chemrxiv.12465371.v1