Learning to Make Generalizable and Diverse Predictions for Retrosynthesis

Chen, Benson; Shen, Tianxiao; Jaakkola, Tommi S.; Barzilay, Regina

doi:10.48550/arxiv.1910.09688

Cited by 24 publications

(35 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…18,41,57 However, they have been shown to make some trivial mistakes 51 and produce reactions with a limited diversity. 40 These models also act as black-boxes; they do not provide the reasoning behind their predictions and are not able to map atoms between the substrates and the product.…”

Section: ■ Related Workmentioning

confidence: 99%

“…Despite the added flexibility and the resulting better coverage of the chemical space, current applications of generative modeling tends to result in a limited set of plausible chemical reactions for a given input. 40 Models such as proposed in Schwaller et al, 17 Karpov et al, 18 Zheng et al 41 belong to the class of sequence to sequence models. They generate a chemical reaction by sequentially outputting individual symbols in the SMILES notation.…”

Section: ■ Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Molecule Edit Graph Attention Network: Modeling Chemical Reactions as Sequences of Graph Edits

Sacha¹,

Błaż²,

Byrski³

et al. 2021

J. Chem. Inf. Model.

100

View full text Add to dashboard Cite

The central challenge in automated synthesis planning is to be able to generate and predict outcomes of a diverse set of chemical reactions. In particular, in many cases, the most likely synthesis pathway cannot be applied due to additional constraints, which requires proposing alternative chemical reactions. With this in mind, we present Molecule Edit Graph Attention Network (MEGAN), an end-to-end encoder−decoder neural model. MEGAN is inspired by models that express a chemical reaction as a sequence of graph edits, akin to the arrow pushing formalism. We extend this model to retrosynthesis prediction (predicting substrates given the product of a chemical reaction) and scale it up to large data sets. We argue that representing the reaction as a sequence of edits enables MEGAN to efficiently explore the space of plausible chemical reactions, maintaining the flexibility of modeling the reaction in an end-to-end fashion and achieving state-of-the-art accuracy in standard benchmarks. Code and trained models are made available online at https://github. com/molecule-one/megan.

show abstract

Section: ■ Related Workmentioning

confidence: 99%

Section: ■ Introductionmentioning

confidence: 99%

Molecule Edit Graph Attention Network: Modeling Chemical Reactions as Sequences of Graph Edits

Sacha¹,

Błaż²,

Byrski³

et al. 2021

J. Chem. Inf. Model.

100

View full text Add to dashboard Cite

show abstract

“…Baselines Template-based GLN (Dai et al, 2019), templatefree G2G (Shi et al, 2020) and RetroXpert (Yan et al, 2020) are primary baselines, which not only achieve state-of-theart performance, but also provide open-source PyTorch code that allows us to verify their effectiveness. To show broad superiority, we also comapre SemiRetro with other baselines, incuding RetroSim (Coley et al, 2017b), NeuralSym (Segler & Waller, 2017), SCROP (Zheng et al, 2019), LV-Transformer (Chen et al, 2019), GraphRetro (Somnath et al, 2021), MEGAN (Sacha et al, 2021), MHNreact (Seidl et al, 2021), and Dual model (Sun et al, 2020). As the retrosynthesis task is quite complex, subtle implementation differences or mistakes may cause critical performance fluctuations.…”

Section: Basic Settingmentioning

confidence: 99%

SemiRetro: Semi-template framework boosts deep retrosynthesis prediction

Gao¹,

Tan²,

Wu³

et al. 2022

Preprint

View full text Add to dashboard Cite

Recently, template-based (TB) and templatefree (TF) molecule graph learning methods have shown promising results to retrosynthesis. TB methods are more accurate using pre-encoded reaction templates, and TF methods are more scalable by decomposing retrosynthesis into subproblems, i.e., center identification and synthon completion. To combine both advantages of TB and TF, we suggest breaking a full-template into several semi-templates and embedding them into the two-step TF framework. Since many semi-templates are reduplicative, the template redundancy can be reduced while the essential chemical knowledge is still preserved to facilitate synthon completion. We call our method SemiRetro, introduce a new GNN layer (DR-GAT) to enhance center identification, and propose a novel self-correcting module to improve semi-template classification. Experimental results show that SemiRetro significantly outperforms both existing TB and TF methods. In scalability, SemiRetro covers 98.9% data using 150 semitemplates, while previous template-based GLN requires 11,647 templates to cover 93.3% data. In top-1 accuracy, SemiRetro exceeds templatefree G2G 4.8% (class known) and 6.0% (class unknown). Besides, SemiRetro has better training efficiency than existing methods.

show abstract

“…22−25 Although template-free models do not require any chemical prior knowledge or reaction templates and exhibit comparable or better accuracy than template-based models, they are likely to provide dissatisfactory results owing to the three types of problems depicted in Figure 1 predicted reactant text representation (candidate 1 in Figure 1) has a grammatical error and cannot be converted into a molecular structure, 21,23 (ii) implausibility: the prediction (candidate 2 in the Figure 1) is grammatically valid but cannot synthesize the desired product, 21 and (iii) lack of diversity: some of the predicted reactant candidates (candidates 3 and 4 in Figure 1) are duplicated, thereby possibly lowering the number of unique suggestions. 24,27 Most of the retrosynthesis studies evaluate the precursor predictions in terms of accuracy, but grammatical validity, chemical plausibility, and diversity are also important. Expanding to multistep retrosynthesis planning, diversity acts as a more important performance indicator than accuracy.…”

Section: ■ Introductionmentioning

confidence: 99%

“…27 To solve the three problems, attempts have been made, such as adding a syntax corrector that can fix a grammatical error in the retrosynthetic prediction 23 or mixture modeling for diversity. 24 However, no studies have tackled all these problems simultaneously.…”

Section: ■ Introductionmentioning

confidence: 99%

Valid, Plausible, and Diverse Retrosynthesis Using Tied Two-Way Transformers with Latent Variables

Kim

Lee

Kwon

et al. 2021

J. Chem. Inf. Model.

View full text Add to dashboard Cite

Retrosynthesis is an essential task in organic chemistry for identifying the synthesis pathways of newly discovered materials, and with the recent advances in deep learning, there have been growing attempts to solve the retrosynthesis problem through transformer models, which are the state-of-the-art in neural machine translation, by converting the problem into a machine translation problem. However, the pure transformer provides unsatisfactory results that lack grammatical validity, chemical plausibility, and diversity in reactant candidates. In this study, we develop tied two-way transformers with latent modeling to solve those problems using cycle consistency checks, parameter sharing, and multinomial latent variables. Experimental results obtained using public and in-house datasets demonstrate that the proposed model improves the retrosynthesis accuracy, grammatical error, and diversity, and qualitative evaluation results verify its ability to suggest valid and plausible results.

show abstract

Learning to Make Generalizable and Diverse Predictions for Retrosynthesis

Cited by 24 publications

References 18 publications

Molecule Edit Graph Attention Network: Modeling Chemical Reactions as Sequences of Graph Edits

Molecule Edit Graph Attention Network: Modeling Chemical Reactions as Sequences of Graph Edits

SemiRetro: Semi-template framework boosts deep retrosynthesis prediction

Valid, Plausible, and Diverse Retrosynthesis Using Tied Two-Way Transformers with Latent Variables

Contact Info

Product

Resources

About