Marwin Segler

Chemical synthesis remains a bottleneck in the discovery and manufacture of functional small molecules ^1–3 . Although artificial intelligence (AI)-assisted synthesis planning has proliferated in recent years, a detailed understanding of its failure modes has not been achieved, and models still struggle with predicting less frequent, yet strategically important reactions, as well as hallucinated, incorrect predictions misaligned with chemists’ expectations ^4–12 . Here, we analyse the failure modes of the present-day AI models and propose RetroChimera: a frontier retrosynthesis model, built on two newly developed components with complementary inductive biases, integrated through a new, learning-based ensembling strategy. Through experiments across several orders of magnitude in data scale, we show RetroChimera outperforms leading baselines, demonstrating robustness outside the training data, as well as the ability to learn from very small numbers of examples per reaction class. Using both pairwise and pointwise setups, we find that organic chemists prefer predictions from RetroChimera over published reference reactions and over other AI models. Finally, we demonstrate zero-shot transfer and fine-tuning on internal datasets from two pharmaceutical companies, showing robust generalization under distribution shift. Our work demonstrates the viability of deep learning for accurate synthesis prediction in increasingly challenging regimes.

Solving computer-aided synthesis planning is essential for enabling fully automated, robot-assisted synthesis workflows and improving the efficiency of drug discovery. A key challenge, however, is bridging the gap between computational route design and practical laboratory execution, particularly the accurate prediction of viable experimental procedures for each synthesis step. In this work, we present QFANG, a scientific reasoning language model capable of generating precise, structured experimental procedures directly from reaction equations, with explicit chain-of-thought reasoning. To develop QFANG, we curated a high-quality dataset comprising 905,990 chemical reactions paired with structured action sequences, extracted and processed from patent literature using large language models. We introduce a Chemistry-Guided Reasoning (CGR) framework that produces chain-of-thought data grounded in chemical knowledge at scale. The model subsequently undergoes supervised fine-tuning to elicit complex chemistry reasoning. Finally, we apply Reinforcement Learning from Verifiable Rewards (RLVR) to further enhance procedural accuracy. Experimental results demonstrate that QFANG outperforms advanced general-purpose reasoning models and nearest-neighbor retrieval baselines, measured by traditional NLP similarity metrics and a chemically aware evaluator using an LLM-as-a-judge. Moreover, QFANG generalizes to certain out-of-domain reaction classes and adapts to variations in laboratory conditions and user-specific constraints. We believe that QFANG's ability to generate high-quality synthesis procedures represents an important step toward bridging the gap between computational synthesis planning and fully automated laboratory synthesis.

Deep learning models for anticipating the products of organic reactions have found many use cases, including validating retrosynthetic pathways and constraining synthesis-based molecular design tools. Despite compelling performance on popular benchmark tasks, strange and erroneous predictions sometimes ensue when using these models in practice. The core issue is that common benchmarks test models in an in-distribution setting, whereas many real-world uses for these models are in out-of-distribution settings and require a greater degree of extrapolation. To better understand how current reaction predictors work in out-of-distribution domains, we report a series of more challenging evaluations of a prototypical SMILES-based deep learning model. First, we illustrate how performance on randomly sampled data sets is overly optimistic compared to performance when generalizing to new patents or new authors. Second, we conduct time splits that evaluate how models perform when tested on reactions published years after those in their training set, mimicking real-world deployment. Finally, we consider extrapolation across reaction classes to reflect what would be required for the discovery of novel reaction types. This panel of tasks can reveal the capabilities and limitations of today's reaction predictors, acting as a crucial first step in the development of tomorrow's next-generation models capable of reaction discovery.

Syntheseus provides reference models and search algorithms as well as metrics to evaluate and improve synthesis planning tools.

Retrosynthesis is the task of planning a series of chemical reactions to create a desired molecule from simpler, buyable molecules. While previous works have proposed algorithms to find optimal solutions for a range of metrics (e.g. shortest, lowest-cost), these works generally overlook the fact that we have imperfect knowledge of the space of possible reactions, meaning plans created by algorithms may not work in a laboratory. In this paper we propose a novel formulation of retrosynthesis in terms of stochastic processes to account for this uncertainty. We then propose a novel greedy algorithm called retro-fallback which maximizes the probability that at least one synthesis plan can be executed in the lab. Using in-silico benchmarks we demonstrate that retro-fallback generally produces better sets of synthesis plans than the popular MCTS and retro* algorithms.

Machine-learning and deep-learning models have been extensively used in cheminformatics to predict molecular properties, to reduce the need for direct measurements, and to accelerate compound prioritization. However, different setups and frameworks and the large number of molecular representations make it difficult to properly evaluate, reproduce, and compare them. Here we present a new PREdictive modeling FramEwoRk for molecular discovery (PREFER), written in Python (version 3.7.7) and based on AutoSklearn (version 0.14.7), that allows comparison between different molecular representations and common machine-learning models. We provide an overview of the design of our framework and show exemplary use cases and results of several representation-model combinations on diverse data sets, both public and in-house. Finally, we discuss the use of PREFER on small data sets. The code of the framework is freely available on GitHub.

Computational techniques, including virtual screening, de novo design, and generative models, play an increasing role in expediting DMTA cycles for modern molecular discovery. However, computationally proposed molecules must be synthetically feasible for laboratory testing. In this perspective, we offer a succinct introduction to the subject, showcase typical workflows to integrate synthesis planning, synthesizability scoring, and molecule generation. Finally, we address limitations and opportunities for future research.

The lead optimization process in drug discovery campaigns is an arduous endeavour where the input of many medicinal chemists is weighed in order to reach a desired molecular property profile. Building the expertise to successfully drive such projects collaboratively is a very time-consuming process that typically spans many years within a chemist's career. In this work we aim to replicate this process by applying artificial intelligence learning-to-rank techniques on feedback that was obtained from 35 chemists at Novartis over the course of several months. We exemplify the usefulness of the learned proxies in routine tasks such as compound prioritization, motif rationalization, and biased \textit{de novo} drug design. Annotated response data is provided, and developed models and code made available through a permissive open-source license.

De novo molecule design algorithms often result in chemically unfeasible or synthetically inaccessible molecules. A natural idea to mitigate this problem is to bias these algorithms toward more easily synthesizable molecules using a proxy score for synthetic accessibility. However, using currently available proxies can still result in highly unrealistic compounds. Here, we propose a novel approach, RetroGNN, to estimate synthesizability. First, we search for routes using synthesis planning software for a large number of random molecules. This information is then used to train a graph neural network to predict the outcome of the synthesis planner given the target molecule, in which the regression task can be used as a synthesizability scorer. We highlight how RetroGNN can be used in generative molecule-discovery pipelines together with other scoring functions. We evaluate our approach on several QSAR-based molecule design benchmarks, for which we find synthesizable molecules with state-of-the-art scores. Compared to the virtual screening of 5 million existing molecules from the ZINC database, using RetroGNNScore with a simple fragment-based de novo design algorithm finds molecules predicted to be more likely to possess the desired activity exponentially faster, while maintaining good druglike properties and being easier to synthesize. Importantly, our deep neural network can successfully filter out hard to synthesize molecules while achieving a 10⁵ times speedup over using retrosynthesis planning software.