Reagent prediction with a molecular transformer improves reaction data quality

Andronov, Mikhail; Voinarovska, Varvara; Andronova, Natalia; Wand, Michael; Clevert, Djork-Arné; Schmidhuber, Jürgen

doi:10.1039/d2sc06798f

Cited by 20 publications

(17 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Foundational Chemistry Model. As demonstrated in previous publications, 90,93,99 the general chemistry model was pretrained from scratch by using the well-known USPTO_MIT mixed augmented database that contains approximately one million unlabeled organic chemical reactions. Briefly, this step was included to increase the model's vocabulary (∼5000 unique tokens) by providing sufficient chemical information in the form of text notation.…”

Section: Resultsmentioning

confidence: 99%

“…The output model containing all of the trained parameters was stored in a directory called the foundational general chemistry model. The molecular Transformer USPTO_MIT Mixed Augmented database , was used to train as well as to evaluate the proposed chemistry model.…”

Section: Materials and Methodsmentioning

confidence: 99%

“…The foundational general chemistry model was developed by following the same strategy described in a previous publication of our group. 90 Briefly, Transformer-type architectures 91 are now commonly used in Natural Language Processing and are pretrained using Self-Supervised Learning on large amounts of text data to create Foundation Models for different human languages. This pretraining is expensive (in terms of computational resources) but gives the model a broad understanding of the language(s), allowing it to adapt to specific downstream tasks quickly and efficiently.…”

Section: Foundational Chemistry Modelmentioning

confidence: 99%

See 2 more Smart Citations

Predicting Antioxidant Synergism via Artificial Intelligence and Benchtop Data

Ayres,

Benavidez,

Varillas

et al. 2023

J. Agric. Food Chem.

View full text Add to dashboard Cite

Lipid oxidation is a major issue affecting products containing unsaturated fatty acids as ingredients or components, leading to the formation of low molecular weight species with diverse functional groups that impart off-odors and off-flavors. Aiming to control this process, antioxidants are commonly added to these products, often deployed as combinations of two or more compounds, a strategy that allows for lowering the amount used while boosting the total antioxidant capacity of the formulation. While this approach allows for minimizing the potential organoleptic and toxic effects of these compounds, predicting how these mixtures of antioxidants will behave has traditionally been one of the most challenging tasks, often leading to simple additive, antagonistic, or synergistic effects. Approaches to understanding these interactions have been predominantly empirically driven but thus far, inefficient and unable to account for the complexity and multifaceted nature of antioxidant responses. To address this current gap in knowledge, we describe the use of an artificial intelligence model based on deep learning architecture to predict the type of interaction (synergistic, additive, and antagonistic) of antioxidant combinations. Here, each mixture was associated with a combination index value (CI) and used as input for our model, which was challenged against a test (n = 140) data set. Despite the encouraging preliminary results, this algorithm failed to provide accurate predictions of oxidation experiments performed in-house using binary mixtures of phenolic antioxidants and a lard sample. To overcome this problem, the AI algorithm was then enhanced with various amounts of experimental data (antioxidant power data assessed by the TBARS assay), demonstrating the importance of having chemically relevant experimental data to enhance the model’s performance and provide suitable predictions with statistical relevance. We believe the proposed method could be used as an auxiliary tool in benchmark analysis routines, offering a novel strategy to enable broader and more rational predictions related to the behavior of antioxidant mixtures.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Materials and Methodsmentioning

confidence: 99%

Section: Foundational Chemistry Modelmentioning

confidence: 99%

See 1 more Smart Citation

Predicting Antioxidant Synergism via Artificial Intelligence and Benchtop Data

Ayres,

Benavidez,

Varillas

et al. 2023

J. Agric. Food Chem.

View full text Add to dashboard Cite

show abstract

“…Subsequently to our preprinted report, a separate study has investigated the performance of the reagent prediction transformer. 36 Most importantly, when tested on unseen molecules, the TTL provides validated disconnections at several possible reactive sites. By contrast, the baseline transformer, trained as reported by Schwaller et al 24 to produce SM directly from P using the unannotated data for training, chooses fewer disconnection points, as exemplified here for the pro-nucleotide 1 (Figure 2d).…”

Section: Triple Transformer Loop (Ttl) For Single-step Retrosynthesismentioning

confidence: 99%

Multistep retrosynthesis combining a disconnection aware triple transformer loop with a route penalty score guided tree search

Kreutter

Reymond

2023

Preprint

View full text Add to dashboard Cite

Computer-aided synthesis planning (CASP) aims to automatically learn organic reactivity from literature and perform retrosynthesis of unseen molecules. CASP systems must learn reactions sufficiently precisely to propose realistic disconnections while avoiding overfitting to leave room for diverse options, and explore possible routes such as to allow short synthetic sequences to emerge. Herein we report an open-source CASP tool proposing original solutions to both challenges. First, we use a triple transformer loop (TTL) predicting starting materials (T1), reagents (T2), and products (T3) to explore various disconnections sites defined by combining exhaustive, template-based and transformer-based tagging procedures. Second, we integrate TTL into a multistep tree search algorithm (TTLA) prioritizing sequences using a route penalty score (RPScore) considering the number of steps, their confidence score, and the simplicity of all intermediates along the route. Our approach favours short synthetic routes to commercial starting materials, as exemplified by retrosynthetic analyses of recently approved drugs.

show abstract

“…Gimadiev et al presented a 4-step protocol for cleaning of molecular structures using data originating from Reaxys, USPTO, and Pistachio (e.g., functional group standardization, valence checking) as well as curation of the reaction transformation (e.g., via reaction balancing or atom mapping), but no further application such as predictive modeling was conducted. Andronov et al published a cleaning pipeline involving atom-mapping, removal of isotope information, and SMILES canonicalization for subsequent training of a transformer model for single-step retrosynthesis . ORDerly took inspiration from these previously published works to develop an open-source cleaning pipeline integrated with ORD, providing numerous reaction task benchmarks that have undergone in silico validation.…”

Section: Introductionmentioning

confidence: 99%

ORDerly: Data Sets and Benchmarks for Chemical Reaction Data

Wigh,

Arrowsmith,

Pomberger

et al. 2024

J. Chem. Inf. Model.

View full text Add to dashboard Cite

Machine learning has the potential to provide tremendous value to life sciences by providing models that aid in the discovery of new molecules and reduce the time for new products to come to market. Chemical reactions play a significant role in these fields, but there is a lack of high-quality open-source chemical reaction data sets for training machine learning models. Herein, we present ORDerly, an open-source Python package for the customizable and reproducible preparation of reaction data stored in accordance with the increasingly popular Open Reaction Database (ORD) schema. We use ORDerly to clean United States patent data stored in ORD and generate data sets for forward prediction, retrosynthesis, as well as the first benchmark for reaction condition prediction. We train neural networks on data sets generated with ORDerly for condition prediction and show that data sets missing key cleaning steps can lead to silently overinflated performance metrics. Additionally, we train transformers for forward and retrosynthesis prediction and demonstrate how non-patent data can be used to evaluate model generalization. By providing a customizable open-source solution for cleaning and preparing large chemical reaction data, ORDerly is poised to push forward the boundaries of machine learning applications in chemistry.

show abstract

Reagent prediction with a molecular transformer improves reaction data quality

Abstract: A molecular transformer predicts reagents for organic reactions. It is also able to replace questionable reagents in reaction data, e.g. USPTO, to enable better product prediction models to be trained on these new data.

Cited by 20 publications

References 46 publications

Predicting Antioxidant Synergism via Artificial Intelligence and Benchtop Data

Predicting Antioxidant Synergism via Artificial Intelligence and Benchtop Data

Multistep retrosynthesis combining a disconnection aware triple transformer loop with a route penalty score guided tree search

ORDerly: Data Sets and Benchmarks for Chemical Reaction Data

Contact Info

Product

Resources

About