Modelling Chemical Reasoning to Predict and Invent Reactions

Challenging Reaction Prediction Models to Generalize to Novel Chemistry

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.

Artificial Intelligence in Retrosynthesis Prediction and its Applications in Medicinal Chemistry

Retrosynthesis is a strategy to analyze the synthetic routes for target molecules in medicinal chemistry. However, traditional retrosynthesis predictions performed by chemists and rule-based expert systems struggle to adapt to the vast chemical space of real-world scenarios. Artificial intelligence (AI) has revolutionized retrosynthesis prediction in recent decades, significantly increasing the accuracy and diversity of predictions for target compounds. Single-step AI-driven retrosynthesis models can be generalized into three types based on their dependence on predefined reaction templates (template-based, semitemplate-based methods, template-free models), with respective advantages and limitations, and common challenges that limit their medicinal chemistry applications. Moreover, there are relatively inadequate multi-step retrosynthesis methods, which lack strong links with single-step methods. Herein, we review the recent advancements in AI applications for retrosynthesis prediction by summarizing related techniques and the landscape of current representative retrosynthesis models and propose feasible solutions to tackle existing problems and outline future directions in this field.

Single-step retrosynthesis prediction via multitask graph representation learning

Inferring appropriate synthesis reaction (i.e., retrosynthesis) routes for newly designed molecules is vital. Recently, computational methods have produced promising single-step retrosynthesis predictions. However, template-based methods are limited by the known synthesis templates; template-free methods are weakly interpretable; and semi template-based methods are deficient with regard to utilizing the associations between chemical entities. To address these issues, this paper leverages the intra-associations between synthons, the inter-associations between synthons and leaving groups (LGs), and the intra-associations between LGs. It develops a multitask graph representation learning model for single-step retrosynthesis prediction (Retro-MTGR) to solve reaction centre deduction and LG identification simultaneously. A comparison with 16 state-of-the-art methods first demonstrates the superiority of Retro-MTGR. Then, its robustness and scalability and the contributions of its crucial components are validated. More importantly, it can determine whether a bond can be a reaction centre and what LGs are appropriate for a given synthon, respectively. The answers reflect underlying chemical synthesis rules, especially opposite electrical properties between chemical entities (e.g., reaction sites, synthons, and LGs). Finally, case studies demonstrate that the retrosynthesis routes inferred by Retro-MTGR are promising for single-step synthesis reactions. The code and data of this study are freely available at https://doi.org/10.5281/zenodo.14346324 .

Computer-Aided Synthesis Planning (CASP) and Machine Learning: Optimizing Chemical Reaction Conditions

Computer-aided synthesis planning (CASP) has garnered increasing attention in light of recent advancements in machine learning models. While the focus is on reverse synthesis or forward outcome prediction, optimizing reaction conditions remains a significant challenge. For datasets with multiple variables, the choice of descriptors and models is pivotal. This selection dictates the effective extraction of conditional features and the achievement of higher prediction accuracy. This review delineates the origins of data in conditional optimization, the criteria for descriptor selection, the response models, and the metrics for outcome evaluation, aiming to acquaint readers with the latest research trends and facilitate more informed research in this domain.

Do Chemformers Dream of Organic Matter? Evaluating a Transformer Model for Multistep Retrosynthesis

Synthesis planning of new pharmaceutical compounds is a well-known bottleneck in modern drug design. Template-free methods, such as transformers, have recently been proposed as an alternative to template-based methods for single-step retrosynthetic predictions. Here, we trained and evaluated a transformer model, called the Chemformer, for retrosynthesis predictions within drug discovery. The proprietary data set used for training comprised ∼18 M reactions from literature, patents, and electronic lab notebooks. Chemformer was evaluated for the purpose of both single-step and multistep retrosynthesis. We found that the single-step performance of Chemformer was especially good on reaction classes common in drug discovery, with most reaction classes showing a top-10 round-trip accuracy above 0.97. Moreover, Chemformer reached a higher round-trip accuracy compared to that of a template-based model. By analyzing multistep retrosynthesis experiments, we observed that Chemformer found synthetic routes, leading to commercial starting materials for 95% of the target compounds, an increase of more than 20% compared to the template-based model on a proprietary compound data set. In addition to this, we discovered that Chemformer suggested novel disconnections corresponding to reaction templates, which are not included in the template-based model. These findings were further supported by a publicly available ChEMBL compound data set. The conclusions drawn from this work allow for the design of a synthesis planning tool where template-based and template-free models work in harmony to optimize retrosynthetic recommendations.

Deep Learning with Geometry-Enhanced Molecular Representation for Augmentation of Large-Scale Docking-Based Virtual Screening

Structure-based virtual screening has been a crucial tool in drug discovery for decades. However, as the chemical space expands, the existing structure-based virtual screening techniques based on molecular docking and scoring struggle to handle billion-entry ultralarge libraries due to the high computational cost. To address this challenge, people have resorted to machine learning techniques to enhance structure-based virtual screening for efficiently exploring the vast chemical space. In those cases, compounds are usually treated as sequential strings or two-dimensional topology graphs, limiting their ability to incorporate three-dimensional structural information for downstream tasks. We herein propose a novel deep learning protocol, GEM-Screen, which utilizes the geometry-enhanced molecular representation of the compounds docking to a specific target and is trained on docking scores of a small fraction of a library through an active learning strategy to approximate the docking outcome for yet nontraining entries. This protocol is applied to virtual screening campaigns against the AmpC and D4 targets, demonstrating that GEM-Screen enriches more than 90% of the hit scaffolds for AmpC in the top 4% of model predictions and more than 80% of the hit scaffolds for D4 in the same top-ranking size of library. GEM-Screen can be used in conjunction with traditional docking programs for docking of only the top-ranked compounds to avoid the exhaustive docking of the whole library, thus allowing for discovering top-scoring compounds from billion-entry libraries in a rapid yet accurate fashion.

Machine Learning Strategies for Reaction Development: Toward the Low-Data Limit

Machine learning models are increasingly being utilized to predict outcomes of organic chemical reactions. A large amount of reaction data is used to train these models, which is in stark contrast to how expert chemists discover and develop new reactions by leveraging information from a small number of relevant transformations. Transfer learning and active learning are two strategies that can operate in low-data situations, which may help fill this gap and promote the use of machine learning for tackling real-world challenges in organic synthesis. This Perspective introduces active and transfer learning and connects these to potential opportunities and directions for further research, especially in the area of prospective development of chemical transformations.

A Review on Artificial Intelligence Enabled Design, Synthesis, and Process Optimization of Chemical Products for Industry 4.0

With the development of Industry 4.0, artificial intelligence (AI) is gaining increasing attention for its performance in solving particularly complex problems in industrial chemistry and chemical engineering. Therefore, this review provides an overview of the application of AI techniques, in particular machine learning, in chemical design, synthesis, and process optimization over the past years. In this review, the focus is on the application of AI for structure-function relationship analysis, synthetic route planning, and automated synthesis. Finally, we discuss the challenges and future of AI in making chemical products.

Predictive chemistry: machine learning for reaction deployment, reaction development, and reaction discovery

The field of predictive chemistry relates to the development of models able to describe how molecules interact and react. It encompasses the long-standing task of computer-aided retrosynthesis, but is far more reaching and ambitious in its goals. In this review, we summarize several areas where predictive chemistry models hold the potential to accelerate the deployment, development, and discovery of organic reactions and advance synthetic chemistry.

Self-Supervised Molecular Pretraining Strategy for Low-Resource Reaction Prediction Scenarios

In the face of low-resource reaction training samples, we construct a chemical platform for addressing small-scale reaction prediction problems. Using a self-supervised pretraining strategy called MAsked Sequence to Sequence (MASS), the Transformer model can absorb the chemical information of about 1 billion molecules and then fine-tune on a small-scale reaction prediction. To further strengthen the predictive performance of our model, we combine MASS with the reaction transfer learning strategy. Here, we show that the average improved accuracies of the Transformer model can reach 14.07, 24.26, 40.31, and 57.69% in predicting the Baeyer-Villiger, Heck, C-C bond formation, and functional group interconversion reaction data sets, respectively, marking an important step to low-resource reaction prediction.

RetroComposer: Composing Templates for Template-Based Retrosynthesis Prediction

The main target of retrosynthesis is to recursively decompose desired molecules into available building blocks. Existing template-based retrosynthesis methods follow a template selection stereotype and suffer from limited training templates, which prevents them from discovering novel reactions. To overcome this limitation, we propose an innovative retrosynthesis prediction framework that can compose novel templates beyond training templates. As far as we know, this is the first method that uses machine learning to compose reaction templates for retrosynthesis prediction. Besides, we propose an effective reactant candidate scoring model that can capture atom-level transformations, which helps our method outperform previous methods on the USPTO-50K dataset. Experimental results show that our method can produce novel templates for 15 USPTO-50K test reactions that are not covered by training templates. We have released our source implementation.

Molecular mechanisms from reaction coordinate graph enabled multidimensional free energies illustrated on water dimer hydrogen bonding

Computing the free energies of molecular mechanisms in multidimensional space relies on combinations of geometrically complex reaction coordinates. We show how a graph theory implementation reduces complexity, and illustrate this on the arrangements of hydrogen bonding of a water dimer. The reaction coordinates and forces are computed using graphs that define the dependencies on the atoms in the Free Energy from Adaptive Reaction Coordinate Forces (FEARCF) library. The library can be interfaced with classical molecular dynamics as well as quantum molecular dynamics packages. Multidimensional interdependent reaction coordinates are constructed to produce complex free energy hypersurfaces. The reaction coordinates are graphed from atomic and molecular components to define points, distances, vectors, angles, planes and combinations thereof. The resultant free energy surfaces that are a function of the distance, angles, planes, and so on, can represent molecular mechanisms in reduced dimensions from the component atomic Cartesian coordinate degrees of freedom. The FEARCF library can be interfaced with any molecular package. Here, we demonstrate the link to NWChem to compute a hyperdimensional DFT (aug‐cc‐pVDZ basis set and X3LYP exchange correlation functionals) free energy space of a water dimer. Analysis of the water dimer free energy hypervolume reveals that while the chain and cyclic hydrogen bonding configurations are located in stable minimum energy wells, the bifurcated hydrogen bond configuration is a gateway to instability and dimer dissociation.

Automated Rational Design of Metal-Organic Polyhedra

Metal-organic polyhedra (MOPs) are hybrid organic-inorganic nanomolecules, whose rational design depends on harmonious consideration of chemical complementarity and spatial compatibility between two or more types of chemical building units (CBUs). In this work, we apply knowledge engineering technology to automate the derivation of MOP formulations based on existing knowledge. For this purpose we have (i) curated relevant MOP and CBU data; (ii) developed an assembly model concept that embeds rules in the MOP construction; (iii) developed an OntoMOPs ontology that defines MOPs and their key properties; (iv) input agents that populate The World Avatar (TWA) knowledge graph; and (v) input agents that, using information from TWA, derive a list of new constructible MOPs. Our result provides rapid and automated instantiation of MOPs in TWA and unveils the immediate chemical space of known MOPs, thus shedding light on new MOP targets for future investigations.

Deep learning driven biosynthetic pathways navigation for natural products with BioNavi-NP

The complete biosynthetic pathways are unknown for most natural products (NPs), it is thus valuable to make computer-aided bio-retrosynthesis predictions. Here, a navigable and user-friendly toolkit, BioNavi-NP, is developed to predict the biosynthetic pathways for both NPs and NP-like compounds. First, a single-step bio-retrosynthesis prediction model is trained using both general organic and biosynthetic reactions through end-to-end transformer neural networks. Based on this model, plausible biosynthetic pathways can be efficiently sampled through an AND-OR tree-based planning algorithm from iterative multi-step bio-retrosynthetic routes. Extensive evaluations reveal that BioNavi-NP can identify biosynthetic pathways for 90.2% of 368 test compounds and recover the reported building blocks as in the test set for 72.8%, 1.7 times more accurate than existing conventional rule-based approaches. The model is further shown to identify biologically plausible pathways for complex NPs collected from the recent literature. The toolkit as well as the curated datasets and learned models are freely available to facilitate the elucidation and reconstruction of the biosynthetic pathways for NPs.

The complete biosynthetic pathway from most natural products (NPs) are unknown. Here, the authors report BioNavi-NP, a computational toolkit for bio-retrosynthetic pathway elucidation or reconstruction for both NPs and NP-like compounds.

Computer-Aided Methods for Molecular Classification

The study aims to analyze the degree of similarity of some molecules belonging to two subgroups of Aminoalkylindoles. After extracting the molecules’ characteristics using Cheminformatics methods, and the computation of the Tanimoto coefficients, dendrograms and heatmaps were built to reveal the degree of similarity of the analyzed drugs. Some atom-pair similarities between the molecules in the same group were detected. The clusters determined by the k-means method divided the Benzoylindoles into two subgroups but kept all the Phenylacetylindoles together in the same set. The activity spectrum of the elements in each group was also analyzed, and similarities have been emphasized. The clustering has been validated using the Kruskal–Wallis test on the series of computed probabilities of the main effects.

Unified Deep Learning Model for Multitask Reaction Predictions with Explanation

There is significant interest and importance to develop robust machine learning models to assist organic chemistry synthesis. Typically, task-specific machine learning models for distinct reaction prediction tasks have been developed. In this work, we develop a unified deep learning model, T5Chem, for a variety of chemical reaction predictions tasks by adapting the "Text-to-Text Transfer Transformer" (T5) framework in natural language processing (NLP). On the basis of self-supervised pretraining with PubChem molecules, the T5Chem model can achieve state-of-the-art performances for four distinct types of task-specific reaction prediction tasks using four different open-source data sets, including reaction type classification on USPTO_TPL, forward reaction prediction on USPTO_MIT, single-step retrosynthesis on USPTO_50k, and reaction yield prediction on high-throughput C-N coupling reactions. Meanwhile, we introduced a new unified multitask reaction prediction data set USPTO_500_MT, which can be used to train and test five different types of reaction tasks, including the above four as well as a new reagent suggestion task. Our results showed that models trained with multiple tasks are more robust and can benefit from mutual learning on related tasks. Furthermore, we demonstrated the use of SHAP (SHapley Additive exPlanations) to explain T5Chem predictions at the functional group level, which provides a way to demystify sequence-based deep learning models in chemistry. T5Chem is accessible through https://yzhang.hpc.nyu.edu/T5Chem.

Improving the performance of models for one-step retrosynthesis through re-ranking

Abstract

Retrosynthesis is at the core of organic chemistry. Recently, the rapid growth of artificial intelligence (AI) has spurred a variety of novel machine learning approaches for data-driven synthesis planning. These methods learn complex patterns from reaction databases in order to predict, for a given product, sets of reactants that can be used to synthesise that product. However, their performance as measured by the top-N accuracy in matching published reaction precedents still leaves room for improvement. This work aims to enhance these models by learning to re-rank their reactant predictions. Specifically, we design and train an energy-based model to re-rank, for each product, the published reaction as the top suggestion and the remaining reactant predictions as lower-ranked. We show that re-ranking can improve one-step models significantly using the standard USPTO-50k benchmark dataset, such as RetroSim, a similarity-based method, from 35.7 to 51.8% top-1 accuracy and NeuralSym, a deep learning method, from 45.7 to 51.3%, and also that re-ranking the union of two models’ suggestions can lead to better performance than either alone. However, the state-of-the-art top-1 accuracy is not improved by this method.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s13321-022-00594-8.

Retrosynthetic reaction pathway prediction through neural machine translation of atomic environments

Designing efficient synthetic routes for a target molecule remains a major challenge in organic synthesis. Atom environments are ideal, stand-alone, chemically meaningful building blocks providing a high-resolution molecular representation. Our approach mimics chemical reasoning, and predicts reactant candidates by learning the changes of atom environments associated with the chemical reaction. Through careful inspection of reactant candidates, we demonstrate atom environments as promising descriptors for studying reaction route prediction and discovery. Here, we present a new single-step retrosynthesis prediction method, viz. RetroTRAE, being free from all SMILES-based translation issues, yields a top-1 accuracy of 58.3% on the USPTO test dataset, and top-1 accuracy reaches to 61.6% with the inclusion of highly similar analogs, outperforming other state-of-the-art neural machine translation-based methods. Our methodology introduces a novel scheme for fragmental and topological descriptors to be used as natural inputs for retrosynthetic prediction tasks.

From Platform to Knowledge Graph: Evolution of Laboratory Automation

High-fidelity computer-aided experimentation is becoming more accessible with the development of computing power and artificial intelligence tools. The advancement of experimental hardware also empowers researchers to reach a level of accuracy that was not possible in the past. Marching toward the next generation of self-driving laboratories, the orchestration of both resources lies at the focal point of autonomous discovery in chemical science. To achieve such a goal, algorithmically accessible data representations and standardized communication protocols are indispensable. In this perspective, we recategorize the recently introduced approach based on Materials Acceleration Platforms into five functional components and discuss recent case studies that focus on the data representation and exchange scheme between different components. Emerging technologies for interoperable data representation and multi-agent systems are also discussed with their recent applications in chemical automation. We hypothesize that knowledge graph technology, orchestrating semantic web technologies and multi-agent systems, will be the driving force to bring data to knowledge, evolving our way of automating the laboratory.

Accelerating Manufacturing for Biomass Conversion via Integrated Process and Bench Digitalization: A Perspective

We present a perspective for accelerating biomass manufacturing via digitalization. We summarize the challenges for manufacturing and identify areas where digitalization can help. A profound potential in using lignocellulosic biomass...

Prediction of Optimal Conditions of Hydrogenation Reaction Using the Likelihood Ranking Approach

The selection of experimental conditions leading to a reasonable yield is an important and essential element for the automated development of a synthesis plan and the subsequent synthesis of the target compound. The classical QSPR approach, requiring one-to-one correspondence between chemical structure and a target property, can be used for optimal reaction conditions prediction only on a limited scale when only one condition component (e.g., catalyst or solvent) is considered. However, a particular reaction can proceed under several different conditions. In this paper, we describe the Likelihood Ranking Model representing an artificial neural network that outputs a list of different conditions ranked according to their suitability to a given chemical transformation. Benchmarking calculations demonstrated that our model outperformed some popular approaches to the theoretical assessment of reaction conditions, such as k Nearest Neighbors, and a recurrent artificial neural network performance prediction of condition components (reagents, solvents, catalysts, and temperature). The ability of the Likelihood Ranking model trained on a hydrogenation reactions dataset, (~42,000 reactions) from Reaxys^® database, to propose conditions that led to the desired product was validated experimentally on a set of three reactions with rich selectivity issues.

Artificial Intelligence, Machine Learning, and Deep Learning in Real-Life Drug Design Cases

The discovery and development of drugs is a long and expensive process with a high attrition rate. Computational drug discovery contributes to ligand discovery and optimization, by using models that describe the properties of ligands and their interactions with biological targets. In recent years, artificial intelligence (AI) has made remarkable modeling progress, driven by new algorithms and by the increase in computing power and storage capacities, which allow the processing of large amounts of data in a short time. This review provides the current state of the art of AI methods applied to drug discovery, with a focus on structure- and ligand-based virtual screening, library design and high-throughput analysis, drug repurposing and drug sensitivity, de novo design, chemical reactions and synthetic accessibility, ADMET, and quantum mechanics.

Chemical data intelligence for sustainable chemistry

This study highlights new opportunities for optimal reaction route selection from large chemical databases brought about by the rapid digitalisation of chemical data. The chemical industry requires a transformation towards more sustainable practices, eliminating its dependencies on fossil fuels and limiting its impact on the environment. However, identifying more sustainable process alternatives is, at present, a cumbersome, manual, iterative process, based on chemical intuition and modelling. We give a perspective on methods for automated discovery and assessment of competitive sustainable reaction routes based on renewable or waste feedstocks. Three key areas of transition are outlined and reviewed based on their state-of-the-art as well as bottlenecks: (i) data, (ii) evaluation metrics, and (iii) decision-making. We elucidate their synergies and interfaces since only together these areas can bring about the most benefit. The field of chemical data intelligence offers the opportunity to identify the inherently more sustainable reaction pathways and to identify opportunities for a circular chemical economy. Our review shows that at present the field of data brings about most bottlenecks, such as data completion and data linkage, but also offers the principal opportunity for advancement.

Chemputation and the Standardization of Chemical Informatics

The explosion in the use of machine learning for automated chemical reaction optimization is gathering pace. However, the lack of a standard architecture that connects the concept of chemical transformations universally to software and hardware provides a barrier to using the results of these optimizations and could cause the loss of relevant data and prevent reactions from being reproducible or unexpected findings verifiable or explainable. In this Perspective, we describe how the development of the field of digital chemistry or chemputation, that is the universal code-enabled control of chemical reactions using a standard language and ontology, will remove these barriers allowing users to focus on the chemistry and plug in algorithms according to the problem space to be explored or unit function to be optimized. We describe a standard hardware (the chemical processing programming architecture-the ChemPU) to encompass all chemical synthesis, an approach which unifies all chemistry automation strategies, from solid-phase peptide synthesis, to HTE flow chemistry platforms, while at the same time establishing a publication standard so that researchers can exchange chemical code (χDL) to ensure reproducibility and interoperability. Not only can a vast range of different chemistries be plugged into the hardware, but the ever-expanding developments in software and algorithms can also be accommodated. These technologies, when combined will allow chemistry, or chemputation, to follow computation-that is the running of code across many different types of capable hardware to get the same result every time with a low error rate.

Deep learning in retrosynthesis planning: datasets, models and tools

In recent years, synthesizing drugs powered by artificial intelligence has brought great convenience to society. Since retrosynthetic analysis occupies an essential position in synthetic chemistry, it has received broad attention from researchers. In this review, we comprehensively summarize the development process of retrosynthesis in the context of deep learning. This review covers all aspects of retrosynthesis, including datasets, models and tools. Specifically, we report representative models from academia, in addition to a detailed description of the available and stable platforms in the industry. We also discuss the disadvantages of the existing models and provide potential future trends, so that more abecedarians will quickly understand and participate in the family of retrosynthesis planning.

Machine Learning in Chemical Product Engineering: The State of the Art and a Guide for Newcomers

Chemical Product Engineering (CPE) is marked by numerous challenges, such as the complexity of the properties–structure–ingredients–process relationship of the different products and the necessity to discover and develop constantly and quickly new molecules and materials with tailor-made properties. In recent years, artificial intelligence (AI) and machine learning (ML) methods have gained increasing attention due to their performance in tackling particularly complex problems in various areas, such as computer vision and natural language processing. As such, they present a specific interest in addressing the complex challenges of CPE. This article provides an updated review of the state of the art regarding the implementation of ML techniques in different types of CPE problems with a particular focus on four specific domains, namely the design and discovery of new molecules and materials, the modeling of processes, the prediction of chemical reactions/retrosynthesis and the support for sensorial analysis. This review is further completed by general guidelines for the selection of an appropriate ML technique given the characteristics of each problem and by a critical discussion of several key issues associated with the development of ML modeling approaches. Accordingly, this paper may serve both the experienced researcher in the field as well as the newcomer.

Rotational Variance-Based Data Augmentation in 3D Graph Convolutional Network

This work proposes the data augmentation by molecular rotation, with consideration that the protein-ligand binding events are rotation-variant. As a proof-of-concept, known active (i. e., 1-labeled) ligands to human β-secretase 1 (BACE-1) are rotated for the generation of 0-labeled data, and the rotation-dependent prediction accuracy of 3D graph convolutional network (3DGCN) is investigated after data augmentation. The data augmentation makes the orientation-recognizing ability of 3DGCN improved significantly in the classification task for BACE-1/ligand binding. Furthermore, the data-augmented 3DGCN has a capability for predicting active ligands from a candidate dataset, via improved performance of orientation recognition, which would be applied to virtual drug screening and discovery.

Transmol: repurposing a language model for molecular generation

Recent advances in convolutional neural networks have inspired the application of deep learning to other disciplines. Even though image processing and natural language processing have turned out to be the most successful, there are many other domains that have also benefited; among them, life sciences in general and chemistry and drug design in particular. In concordance with this observation, from 2018 the scientific community has seen a surge of methodologies related to the generation of diverse molecular libraries using machine learning. However to date, attention mechanisms have not been employed for the problem of de novo molecular generation. Here we employ a variant of transformers, an architecture recently developed for natural language processing, for this purpose. Our results indicate that the adapted Transmol model is indeed applicable for the task of generating molecular libraries and leads to statistically significant increases in some of the core metrics of the MOSES benchmark. The presented model can be tuned to either input-guided or diversity-driven generation modes by applying a standard one-seed and a novel two-seed approach, respectively. Accordingly, the one-seed approach is best suited for the targeted generation of focused libraries composed of close analogues of the seed structure, while the two-seeds approach allows us to dive deeper into under-explored regions of the chemical space by attempting to generate the molecules that resemble both seeds. To gain more insights about the scope of the one-seed approach, we devised a new validation workflow that involves the recreation of known ligands for an important biological target vitamin D receptor. To further benefit the chemical community, the Transmol algorithm has been incorporated into our cheML.io web database of ML-generated molecules as a second generation on-demand methodology.

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

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.

Graph transformation for enzymatic mechanisms

MOTIVATION

The design of enzymes is as challenging as it is consequential for making chemical synthesis in medical and industrial applications more efficient, cost-effective and environmentally friendly. While several aspects of this complex problem are computationally assisted, the drafting of catalytic mechanisms, i.e. the specification of the chemical steps-and hence intermediate states-that the enzyme is meant to implement, is largely left to human expertise. The ability to capture specific chemistries of multistep catalysis in a fashion that enables its computational construction and design is therefore highly desirable and would equally impact the elucidation of existing enzymatic reactions whose mechanisms are unknown.

RESULTS

We use the mathematical framework of graph transformation to express the distinction between rules and reactions in chemistry. We derive about 1000 rules for amino acid side chain chemistry from the M-CSA database, a curated repository of enzymatic mechanisms. Using graph transformation, we are able to propose hundreds of hypothetical catalytic mechanisms for a large number of unrelated reactions in the Rhea database. We analyze these mechanisms to find that they combine in chemically sound fashion individual steps from a variety of known multistep mechanisms, showing that plausible novel mechanisms for catalysis can be constructed computationally.

AVAILABILITY AND IMPLEMENTATION

The source code of the initial prototype of our approach is available at https://github.com/Nojgaard/mechsearch.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Applications of artificial intelligence to drug design and discovery in the big data era: a comprehensive review

Artificial intelligence (AI) renders cutting-edge applications in diverse sectors of society. Due to substantial progress in high-performance computing, the development of superior algorithms, and the accumulation of huge biological and chemical data, computer-assisted drug design technology is playing a key role in drug discovery with its advantages of high efficiency, fast speed, and low cost. Over recent years, due to continuous progress in machine learning (ML) algorithms, AI has been extensively employed in various drug discovery stages. Very recently, drug design and discovery have entered the big data era. ML algorithms have progressively developed into a deep learning technique with potent generalization capability and more effectual big data handling, which further promotes the integration of AI technology and computer-assisted drug discovery technology, hence accelerating the design and discovery of the newest drugs. This review mainly summarizes the application progression of AI technology in the drug discovery process, and explores and compares its advantages over conventional methods. The challenges and limitations of AI in drug design and discovery have also been discussed.

Learning graph representations of biochemical networks and its application to enzymatic link prediction

Motivation

The complete characterization of enzymatic activities between molecules remains incomplete, hindering biological engineering and limiting biological discovery. We develop in this work a technique, enzymatic link prediction (ELP), for predicting the likelihood of an enzymatic transformation between two molecules. ELP models enzymatic reactions cataloged in the KEGG database as a graph. ELP is innovative over prior works in using graph embedding to learn molecular representations that capture not only molecular and enzymatic attributes but also graph connectivity.

Results

We explore transductive (test nodes included in the training graph) and inductive (test nodes not part of the training graph) learning models. We show that ELP achieves high AUC when learning node embeddings using both graph connectivity and node attributes. Further, we show that graph embedding improves link prediction by 30% in area under curve over fingerprint-based similarity approaches and by 8% over support vector machines. We compare ELP against rule-based methods. We also evaluate ELP for predicting links in pathway maps and for reconstruction of edges in reaction networks of four common gut microbiota phyla: actinobacteria, bacteroidetes, firmicutes and proteobacteria. To emphasize the importance of graph embedding in the context of biochemical networks, we illustrate how graph embedding can guide visualization.

Availability and implementation

The code and datasets are available through https://github.com/HassounLab/ELP.

An effective self-supervised framework for learning expressive molecular global representations to drug discovery

How to produce expressive molecular representations is a fundamental challenge in artificial intelligence-driven drug discovery. Graph neural network (GNN) has emerged as a powerful technique for modeling molecular data. However, previous supervised approaches usually suffer from the scarcity of labeled data and poor generalization capability. Here, we propose a novel molecular pre-training graph-based deep learning framework, named MPG, that learns molecular representations from large-scale unlabeled molecules. In MPG, we proposed a powerful GNN for modelling molecular graph named MolGNet, and designed an effective self-supervised strategy for pre-training the model at both the node and graph-level. After pre-training on 11 million unlabeled molecules, we revealed that MolGNet can capture valuable chemical insights to produce interpretable representation. The pre-trained MolGNet can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of drug discovery tasks, including molecular properties prediction, drug-drug interaction and drug-target interaction, on 14 benchmark datasets. The pre-trained MolGNet in MPG has the potential to become an advanced molecular encoder in the drug discovery pipeline.

Organic reactivity from mechanism to machine learning

As more data are introduced in the building of models of chemical reactivity, the mechanistic component can be reduced until 'big data' applications are reached. These methods no longer depend on underlying mechanistic hypotheses, potentially learning them implicitly through extensive data training. Reactivity models often focus on reaction barriers, but can also be trained to directly predict lab-relevant properties, such as yields or conditions. Calculations with a quantum-mechanical component are still preferred for quantitative predictions of reactivity. Although big data applications tend to be more qualitative, they have the advantage to be broadly applied to different kinds of reactions. There is a continuum of methods in between these extremes, such as methods that use quantum-derived data or descriptors in machine learning models. Here, we present an overview of the recent machine learning applications in the field of chemical reactivity from a mechanistic perspective. Starting with a summary of how reactivity questions are addressed by quantum-mechanical methods, we discuss methods that augment or replace quantum-based modelling with faster alternatives relying on machine learning.

Nitromethane Decomposition via Automated Reaction Discovery and an Ab Initio Corrected Kinetic Model

We explore the systematic construction of kinetic models from in silico reaction data for the decomposition of nitromethane. Our models are constructed in a computationally affordable manner by using reactions discovered through accelerated molecular dynamics simulations using the ReaxFF reactive force field. The reaction paths are then optimized to determine reaction rate parameters. We introduce a reaction barrier correction scheme that combines accurate thermochemical data from density functional theory with ReaxFF minimal energy paths. We validate our models across different thermodynamic regimes, showing predictions of gas phase CO and NO concentrations and high-pressure induction times that are similar to experimental data. The kinetic models are analyzed to find fundamental decomposition reactions in different thermodynamic regimes.

Discovery of novel chemical reactions by deep generative recurrent neural network

The "creativity" of Artificial Intelligence (AI) in terms of generating de novo molecular structures opened a novel paradigm in compound design, weaknesses (stability & feasibility issues of such structures) notwithstanding. Here we show that "creative" AI may be as successfully taught to enumerate novel chemical reactions that are stoichiometrically coherent. Furthermore, when coupled to reaction space cartography, de novo reaction design may be focused on the desired reaction class. A sequence-to-sequence autoencoder with bidirectional Long Short-Term Memory layers was trained on on-purpose developed "SMILES/CGR" strings, encoding reactions of the USPTO database. The autoencoder latent space was visualized on a generative topographic map. Novel latent space points were sampled around a map area populated by Suzuki reactions and decoded to corresponding reactions. These can be critically analyzed by the expert, cleaned of irrelevant functional groups and eventually experimentally attempted, herewith enlarging the synthetic purpose of popular synthetic pathways.

Fuzzy knowledge graph system for artificial intelligence-based smart education

Fuzzy knowledge graph system is a semantic network that reveals the relationships between entities, and a tool or methodology that can formally describe things in the real world and their relationships. Smart education is an educational concept or model that uses advanced information technology to build a smart environment, integrates theory and practice to build an educational framework for information age, and provides paths to practice it. Artificial intelligence (AI) is a comprehensive discipline developed by the interpenetration of computer science, cybernetics, information theory, linguistics, neurophysiology and other disciplines, which is a direction for the development of information technology in the future. On the basis of summarizing and analyzing of previous research works, this paper expounded the research status and significance of AI technology, elaborated the development background, current status and future challenges of the construction and application of fuzzy knowledge graph system for smart education, introduced the methods and principles of data acquisition methods and digitalized apprenticeship, realized the process design, information extraction, entity recognition and relationship mining of smart education, constructed a systematic framework for fuzzy knowledge graph, and analyzed the high-quality resources sharing and personalized service of AI-assisted smart education, discussed automatic knowledge acquisition and fusion of fuzzy knowledge graph, performed co-occurrence relationship analysis, and finally conducted application case analysis. The results show that the smart education knowledge graph for AI-assisted smart education can integrate teaching experience and domain knowledge of discipline experts, enhance explainable and robust machine intelligence for AI-assisted smart education, and provide data-driven and knowledge-driven information processing methods; it can also discover the analysis hotspots and main content of research objects through clustering of high-frequency topic words, reveal the corresponding research structure in depth, and then systematically explore its research dimensions, subject background and theoretical basis.

Combined Graph/Relational Database Management System for Calculated Chemical Reaction Pathway Data

Presently, quantum chemical calculations are widely used to generate extensive data sets for machine learning applications; however, generally, these sets only include information on equilibrium structures and some close conformers. Exploration of potential energy surfaces provides important information on ground and transition states, but analysis of such data is complicated due to the number of possible reaction pathways. Here, we present RePathDB, a database system for managing 3D structural data for both ground and transition states resulting from quantum chemical calculations. Our tool allows one to store, assemble, and analyze reaction pathway data. It combines relational database CGR DB for handling compounds and reactions as molecular graphs with a graph database architecture for pathway analysis by graph algorithms. Original condensed graph of reaction technology is used to store any chemical reaction as a single graph.

Substructure-based neural machine translation for retrosynthetic prediction

With the rapid improvement of machine translation approaches, neural machine translation has started to play an important role in retrosynthesis planning, which finds reasonable synthetic pathways for a target molecule. Previous studies showed that utilizing the sequence-to-sequence frameworks of neural machine translation is a promising approach to tackle the retrosynthetic planning problem. In this work, we recast the retrosynthetic planning problem as a language translation problem using a template-free sequence-to-sequence model. The model is trained in an end-to-end and a fully data-driven fashion. Unlike previous models translating the SMILES strings of reactants and products, we introduced a new way of representing a chemical reaction based on molecular fragments. It is demonstrated that the new approach yields better prediction results than current state-of-the-art computational methods. The new approach resolves the major drawbacks of existing retrosynthetic methods such as generating invalid SMILES strings. Specifically, our approach predicts highly similar reactant molecules with an accuracy of 57.7%. In addition, our method yields more robust predictions than existing methods.

Big data and machine learning for materials science

Herein, we review aspects of leading-edge research and innovation in materials science that exploit big data and machine learning (ML), two computer science concepts that combine to yield computational intelligence. ML can accelerate the solution of intricate chemical problems and even solve problems that otherwise would not be tractable. However, the potential benefits of ML come at the cost of big data production; that is, the algorithms demand large volumes of data of various natures and from different sources, from material properties to sensor data. In the survey, we propose a roadmap for future developments with emphasis on computer-aided discovery of new materials and analysis of chemical sensing compounds, both prominent research fields for ML in the context of materials science. In addition to providing an overview of recent advances, we elaborate upon the conceptual and practical limitations of big data and ML applied to materials science, outlining processes, discussing pitfalls, and reviewing cases of success and failure.

Data augmentation and transfer learning strategies for reaction prediction in low chemical data regimes

An effective and rapid deep learning method to predict chemical reactions contributes to the research and development of organic chemistry and drug discovery.

Artificial intelligence and automation in computer aided synthesis planning

In this perspective we deal with questions pertaining to the development of synthesis planning technologies over the course of recent years.

Deep inverse reinforcement learning for structural evolution of small molecules

The size and quality of chemical libraries to the drug discovery pipeline are crucial for developing new drugs or repurposing existing drugs. Existing techniques such as combinatorial organic synthesis and high-throughput screening usually make the process extraordinarily tough and complicated since the search space of synthetically feasible drugs is exorbitantly huge. While reinforcement learning has been mostly exploited in the literature for generating novel compounds, the requirement of designing a reward function that succinctly represents the learning objective could prove daunting in certain complex domains. Generative adversarial network-based methods also mostly discard the discriminator after training and could be hard to train. In this study, we propose a framework for training a compound generator and learn a transferable reward function based on the entropy maximization inverse reinforcement learning (IRL) paradigm. We show from our experiments that the IRL route offers a rational alternative for generating chemical compounds in domains where reward function engineering may be less appealing or impossible while data exhibiting the desired objective is readily available.

Current and Future Roles of Artificial Intelligence in Medicinal Chemistry Synthesis

Artificial intelligence and machine learning have demonstrated their potential role in predictive chemistry and synthetic planning of small molecules; there are at least a few reports of companies employing in silico synthetic planning into their overall approach to accessing target molecules. A data-driven synthesis planning program is one component being developed and evaluated by the Machine Learning for Pharmaceutical Discovery and Synthesis (MLPDS) consortium, comprising MIT and 13 chemical and pharmaceutical company members. Together, we wrote this perspective to share how we think predictive models can be integrated into medicinal chemistry synthesis workflows, how they are currently used within MLPDS member companies, and the outlook for this field.