Exploring Synthesizable Chemical Space with Iterative Pathway Refinements

A well-known pitfall of molecular generative models is that they are not guaranteed to generate synthesizable molecules. Existing solutions for this problem often struggle to effectively navigate exponentially large combinatorial space of synthesizable molecules and suffer from poor coverage. To address this problem, we introduce ReaSyn, an iterative generative pathway refinement framework that obtains synthesizable analogs to input molecules by projecting them onto synthesizable space. Specifically, we propose a simple synthetic pathway representation that allows for generating pathways in both bottom-up and topdown traversal of synthetic trees. We design ReaSyn so that both bottom-up and top-down pathways can be sampled with a single unified autoregressive model. ReaSyn can thus iteratively refine subtrees of generated synthetic trees in a bidirectional manner. Further, we introduce a discrete flow model that refines the generated pathway at the entire pathway level with edit operations: insertion, deletion, and substitution. The iterative refinement cycle of (1) bottom-up decoding, (2) top-down decoding, and (3) holistic editing constitutes a powerful pathway reasoning strategy, allowing the model to explore the vast space of synthesizable molecules. Experimentally, ReaSyn achieves the highest reconstruction rate and pathway diversity in synthesizable molecule reconstruction and the highest optimization performance in synthesizable goal-directed molecular optimization, and significantly outperforms previous synthesizable projection methods in synthesizable hit expansion. These results highlight ReaSyn's superior ability to navigate combinatorially-large synthesizable chemical space. Our code is available at https://github.com/NVIDIA-Digital-Bio/ReaSyn.

• Work during an internship at NVIDIA. † Equal advising.

• We introduce a simple synthetic pathway representation that enables traversal of synthesis trees in both bottom-up and top-down directions.

• We propose a new bidirectional search strategy in synthetic pathway space that unifies bottom-up and top-down sampling.

• We propose Edit Bridge, a novel discrete flow that bridges between the learned distribution in a generative model and the data distribution through edit operations.

• We propose ReaSyn, a framework that integrates bottom-up decoding, top-down decoding, and holistic editing to establish a multi-view pathway generation method for synthesizable projection.

• We demonstrate the effectiveness and versatility of ReaSyn in various synthesizable molecule generation and optimization tasks through extensive experiments.

Synthesizable molecule design. Molecular synthesizability is a vital problem in drug discovery and therefore has received a lot of attention. In synthesizable molecule design, the synthesizable chemical space is defined by sets of reaction rules and building blocks, and the design space is constrained to this