DECIMER.ai: an open platform for automated optical chemical structure identification, segmentation and recognition in scientific publications

The number of publications describing chemical structures has increased steadily over the last decades. However, the majority of published chemical information is currently not available in machine-readable form in public databases. It remains a challenge to automate the process of information extraction in a way that requires less manual intervention - especially the mining of chemical structure depictions. As an open-source platform that leverages recent advancements in deep learning, computer vision, and natural language processing, DECIMER.ai (Deep lEarning for Chemical IMagE Recognition) strives to automatically segment, classify, and translate chemical structure depictions from the printed literature. The segmentation and classification tools are the only openly available packages of their kind, and the optical chemical structure recognition (OCSR) core application yields outstanding performance on all benchmark datasets. The source code, the trained models and the datasets developed in this work have been published under permissive licences. An instance of the DECIMER web application is available at https://decimer.ai .

Open data and algorithms for open science in AI-driven molecular informatics

Recent years have seen a sharp increase in the development of deep learning and artificial intelligence-based molecular informatics. There has been a growing interest in applying deep learning to several subfields, including the digital transformation of synthetic chemistry, extraction of chemical information from the scientific literature, and AI in natural product-based drug discovery. The application of AI to molecular informatics is still constrained by the fact that most of the data used for training and testing deep learning models are not available as FAIR and open data. As open science practices continue to grow in popularity, initiatives which support FAIR and open data as well as open-source software have emerged. It is becoming increasingly important for researchers in the field of molecular informatics to embrace open science and to submit data and software in open repositories. With the advent of open-source deep learning frameworks and cloud computing platforms, academic researchers are now able to deploy and test their own deep learning models with ease. With the development of new and faster hardware for deep learning and the increasing number of initiatives towards digital research data management infrastructures, as well as a culture promoting open data, open source, and open science, AI-driven molecular informatics will continue to grow. This review examines the current state of open data and open algorithms in molecular informatics, as well as ways in which they could be improved in future.

Scaffold Generator: a Java library implementing molecular scaffold functionalities in the Chemistry Development Kit (CDK)

The concept of molecular scaffolds as defining core structures of organic molecules is utilised in many areas of chemistry and cheminformatics, e.g. drug design, chemical classification, or the analysis of high-throughput screening data. Here, we present Scaffold Generator, a comprehensive open library for the generation, handling, and display of molecular scaffolds, scaffold trees and networks. The new library is based on the Chemistry Development Kit (CDK) and highly customisable through multiple settings, e.g. five different structural framework definitions are available. For display of scaffold hierarchies, the open GraphStream Java library is utilised. Performance snapshots with natural products (NP) from the COCONUT (COlleCtion of Open Natural prodUcTs) database and drug molecules from DrugBank are reported. The generation of a scaffold network from more than 450,000 NP can be achieved within a single day.

DECIMER-hand-drawn molecule images dataset

The translation of images of chemical structures into machine-readable representations of the depicted molecules is known as optical chemical structure recognition (OCSR). There has been a lot of progress over the last three decades in this field, but the development of systems for the recognition of complex hand-drawn structure depictions is still at the beginning. Currently, there is no data for the systematic evaluation of OCSR methods on hand-drawn structures available. Here we present DECIMER — Hand-drawn molecule images, a standardised, openly available benchmark dataset of 5088 hand-drawn depictions of diversely picked chemical structures. Every structure depiction in the dataset is mapped to a machine-readable representation of the underlying molecule. The dataset is openly available and published under the CC-BY 4.0 licence which applies very few limitations. We hope that it will contribute to the further development of the field.

RanDepict: Random chemical structure depiction generator

The development of deep learning-based optical chemical structure recognition (OCSR) systems has led to a need for datasets of chemical structure depictions. The diversity of the features in the training data is an important factor for the generation of deep learning systems that generalise well and are not overfit to a specific type of input. In the case of chemical structure depictions, these features are defined by the depiction parameters such as bond length, line thickness, label font style and many others. Here we present RanDepict, a toolkit for the creation of diverse sets of chemical structure depictions. The diversity of the image features is generated by making use of all available depiction parameters in the depiction functionalities of the CDK, RDKit, and Indigo. Furthermore, there is the option to enhance and augment the image with features such as curved arrows, chemical labels around the structure, or other kinds of distortions. Using depiction feature fingerprints, RanDepict ensures diversely picked image features. Here, the depiction and augmentation features are summarised in binary vectors and the MaxMin algorithm is used to pick diverse samples out of all valid options. By making all resources described herein publicly available, we hope to contribute to the development of deep learning-based OCSR systems.

DECIMER 1.0: deep learning for chemical image recognition using transformers

The amount of data available on chemical structures and their properties has increased steadily over the past decades. In particular, articles published before the mid-1990 are available only in printed or scanned form. The extraction and storage of data from those articles in a publicly accessible database are desirable, but doing this manually is a slow and error-prone process. In order to extract chemical structure depictions and convert them into a computer-readable format, Optical Chemical Structure Recognition (OCSR) tools were developed where the best performing OCSR tools are mostly rule-based. The DECIMER (Deep lEarning for Chemical ImagE Recognition) project was launched to address the OCSR problem with the latest computational intelligence methods to provide an automated open-source software solution. Various current deep learning approaches were explored to seek a best-fitting solution to the problem. In a preliminary communication, we outlined the prospect of being able to predict SMILES encodings of chemical structure depictions with about 90% accuracy using a dataset of 50-100 million molecules. In this article, the new DECIMER model is presented, a transformer-based network, which can predict SMILES with above 96% accuracy from depictions of chemical structures without stereochemical information and above 89% accuracy for depictions with stereochemical information.

Quantitative Estimation of Cyclotide-Induced Bilayer Membrane Disruption by Lipid Extraction with Mesoscopic Simulation

Cyclotide-induced membrane disruption is studied at the microsecond timescale by dissipative particle dynamics to quantitatively estimate a kinetic rate constant for membrane lipid extraction with a ″sandwich″ interaction model where two bilayer membranes enclose a cyclotide/water compartment. The obtained bioactivity trends for cyclotides Kalata B1, Cycloviolacin O2, and selected mutants with different membrane types are in agreement with experimental findings: For all membranes investigated, Cycloviolacin O2 shows a higher lipid extraction activity than Kalata B1. The presence of cholesterol leads to a decreased cyclotide activity compared to cholesterol-free membranes. Phosphoethanolamine-rich membranes exhibit an increased membrane disruption. A cyclotide's ″hydrophobic patch″ surface area is important for its bioactivity. A replacement of or with charged amino acid residues may lead to super-mutants with above-native activity but without simple charge-activity patterns. Cyclotide mixtures show linearly additive bioactivities without significant sub- or over-additive effects. The proposed method can be applied as a fast and easy-to-use tool for exploring structure-activity relationships of cyclotide/membrane systems: With the open software provided, the rate constant of a single cyclotide/membrane system can be determined in about 1 day by a scientific end-user without programming skills.

STOUT: SMILES to IUPAC names using neural machine translation

Chemical compounds can be identified through a graphical depiction, a suitable string representation, or a chemical name. A universally accepted naming scheme for chemistry was established by the International Union of Pure and Applied Chemistry (IUPAC) based on a set of rules. Due to the complexity of this ruleset a correct chemical name assignment remains challenging for human beings and there are only a few rule-based cheminformatics toolkits available that support this task in an automated manner. Here we present STOUT (SMILES-TO-IUPAC-name translator), a deep-learning neural machine translation approach to generate the IUPAC name for a given molecule from its SMILES string as well as the reverse translation, i.e. predicting the SMILES string from the IUPAC name. In both cases, the system is able to predict with an average BLEU score of about 90% and a Tanimoto similarity index of more than 0.9. Also incorrect predictions show a remarkable similarity between true and predicted compounds.

DECIMER-Segmentation: Automated extraction of chemical structure depictions from scientific literature

Chemistry looks back at many decades of publications on chemical compounds, their structures and properties, in scientific articles. Liberating this knowledge (semi-)automatically and making it available to the world in open-access databases is a current challenge. Apart from mining textual information, Optical Chemical Structure Recognition (OCSR), the translation of an image of a chemical structure into a machine-readable representation, is part of this workflow. As the OCSR process requires an image containing a chemical structure, there is a need for a publicly available tool that automatically recognizes and segments chemical structure depictions from scientific publications. This is especially important for older documents which are only available as scanned pages. Here, we present DECIMER (Deep lEarning for Chemical IMagE Recognition) Segmentation, the first open-source, deep learning-based tool for automated recognition and segmentation of chemical structures from the scientific literature. The workflow is divided into two main stages. During the detection step, a deep learning model recognizes chemical structure depictions and creates masks which define their positions on the input page. Subsequently, potentially incomplete masks are expanded in a post-processing workflow. The performance of DECIMER Segmentation has been manually evaluated on three sets of publications from different publishers. The approach operates on bitmap images of journal pages to be applicable also to older articles before the introduction of vector images in PDFs. By making the source code and the trained model publicly available, we hope to contribute to the development of comprehensive chemical data extraction workflows. In order to facilitate access to DECIMER Segmentation, we also developed a web application. The web application, available at https://decimer.ai, lets the user upload a pdf file and retrieve the segmented structure depictions.

Too sweet: cheminformatics for deglycosylation in natural products

Sugar units in natural products are pharmacokinetically important but often redundant and therefore obstructing the study of the structure and function of the aglycon. Therefore, it is recommended to remove the sugars before a theoretical or experimental study of a molecule. Deglycogenases, enzymes that specialized in sugar removal from small molecules, are often used in laboratories to perform this task. However, there is no standardized computational procedure to perform this task in silico. In this work, we present a systematic approach for in silico removal of ring and linear sugars from molecular structures. Particular attention is given to molecules of biological origin and to their structural specificities. This approach is made available in two forms, through a free and open web application and as standalone open-source software.

DECIMER: towards deep learning for chemical image recognition

The automatic recognition of chemical structure diagrams from the literature is an indispensable component of workflows to re-discover information about chemicals and to make it available in open-access databases. Here we report preliminary findings in our development of Deep lEarning for Chemical ImagE Recognition (DECIMER), a deep learning method based on existing show-and-tell deep neural networks, which makes very few assumptions about the structure of the underlying problem. It translates a bitmap image of a molecule, as found in publications, into a SMILES. The training state reported here does not yet rival the performance of existing traditional approaches, but we present evidence that our method will reach a comparable detection power with sufficient training time. Training success of DECIMER depends on the input data representation: DeepSMILES are superior over SMILES and we have a preliminary indication that the recently reported SELFIES outperform DeepSMILES. An extrapolation of our results towards larger training data sizes suggests that we might be able to achieve near-accurate prediction with 50 to 100 million training structures. This work is entirely based on open-source software and open data and is available to the general public for any purpose.

MFsim-an open Java all-in-one rich-client simulation environment for mesoscopic simulation

MFsim is an open Java all-in-one rich-client computing environment for mesoscopic simulation with Jdpd as its default simulation kernel for Molecular Fragment (Dissipative Particle) Dynamics. The new environment comprises the complete preparation-simulation-evaluation triad of a mesoscopic simulation task and especially enables biomolecular simulation tasks with peptides and proteins. Productive highlights are a SPICES molecular structure editor, a PDB-to-SPICES parser for particle-based peptide/protein representations, a support of polymer definitions, a compartment editor for complex simulation box start configurations, interactive and flexible simulation box views including analytics, simulation movie generation or animated diagrams. As an open project, MFsim allows for customized extensions for different fields of research.

ErtlFunctionalGroupsFinder: automated rule-based functional group detection with the Chemistry Development Kit (CDK)

The Ertl algorithm for automated functional groups (FG) detection and extraction of organic molecules is implemented on the basis of the Chemistry Development Kit (CDK). A distinct impact of the chosen CDK aromaticity model is demonstrated by an FG analysis of the ChEMBL database compounds. The average performance of less than a millisecond for a single-molecule FG extraction allows for fast processing of even large compound databases.

SPICES: a particle-based molecular structure line notation and support library for mesoscopic simulation

Simplified Particle Input ConnEction Specification (SPICES) is a particle-based molecular structure representation derived from straightforward simplifications of the atom-based SMILES line notation. It aims at supporting tedious and error-prone molecular structure definitions for particle-based mesoscopic simulation techniques like Dissipative Particle Dynamics by allowing for an interplay of different molecular encoding levels that range from topological line notations and corresponding particle-graph visualizations to 3D structures with support of their spatial mapping into a simulation box. An open Java library for SPICES structure handling and mesoscopic simulation support in combination with an open Java Graphical User Interface viewer application for visual topological inspection of SPICES definitions are provided.

Jdpd: an open java simulation kernel for molecular fragment dissipative particle dynamics

Jdpd is an open Java simulation kernel for Molecular Fragment Dissipative Particle Dynamics with parallelizable force calculation, efficient caching options and fast property calculations. It is characterized by an interface and factory-pattern driven design for simple code changes and may help to avoid problems of polyglot programming. Detailed input/output communication, parallelization and process control as well as internal logging capabilities for debugging purposes are supported. The new kernel may be utilized in different simulation environments ranging from flexible scripting solutions up to fully integrated “all-in-one” simulation systems.

A molecular fragment cheminformatics roadmap for mesoscopic simulation

Background

Mesoscopic simulation studies the structure, dynamics and properties of large molecular ensembles with millions of atoms: Its basic interacting units (beads) are no longer the nuclei and electrons of quantum chemical ab-initio calculations or the atom types of molecular mechanics but molecular fragments, molecules or even larger molecular entities. For its simulation setup and output a mesoscopic simulation kernel software uses abstract matrix (array) representations for bead topology and connectivity. Therefore a pure kernel-based mesoscopic simulation task is a tedious, time-consuming and error-prone venture that limits its practical use and application. A consequent cheminformatics approach tackles these problems and provides solutions for a considerably enhanced accessibility. This study aims at outlining a complete cheminformatics roadmap that frames a mesoscopic Molecular Fragment Dynamics (MFD) simulation kernel to allow its efficient use and practical application.

Results

The molecular fragment cheminformatics roadmap consists of four consecutive building blocks: An adequate fragment structure representation (1), defined operations on these fragment structures (2), the description of compartments with defined compositions and structural alignments (3), and the graphical setup and analysis of a whole simulation box (4). The basis of the cheminformatics approach (i.e. building block 1) is a SMILES-like line notation (denoted fSMILES) with connected molecular fragments to represent a molecular structure. The fSMILES notation and the following concepts and methods for building blocks 2-4 are outlined with examples and practical usage scenarios. It is shown that the requirements of the roadmap may be partly covered by already existing open-source cheminformatics software.

Conclusions

Mesoscopic simulation techniques like MFD may be considerably alleviated and broadened for practical use with a consequent cheminformatics layer that successfully tackles its setup subtleties and conceptual usage hurdles. Molecular Fragment Cheminformatics may be regarded as a crucial accelerator to propagate MFD and similar mesoscopic simulation techniques in the molecular sciences.Graphical abstract

Molecular fragment dynamics study on the water-air interface behavior of non-ionic polyoxyethylene alkyl ether surfactants

Molecular fragment dynamics (MFD) is a mesoscopic simulation technique based on dissipative particle dynamics (DPD). MFD simulations of the self-aggregation of the polyoxyethylene alkyl ether surfactants C6E6, C10E6, C12E6 and C16E6 at the water-air surface lead to equilibrium nanoscale structures and computationally determined surface tensions which are in agreement with experimental data for different surfactant concentrations. Thus, molecular fragment dynamics is a well-suited predictive technique to study the behavior of new surfactant systems.

New developments on the cheminformatics open workflow environment CDK-Taverna

BACKGROUND

The computational processing and analysis of small molecules is at heart of cheminformatics and structural bioinformatics and their application in e.g. metabolomics or drug discovery. Pipelining or workflow tools allow for the Lego™-like, graphical assembly of I/O modules and algorithms into a complex workflow which can be easily deployed, modified and tested without the hassle of implementing it into a monolithic application. The CDK-Taverna project aims at building a free open-source cheminformatics pipelining solution through combination of different open-source projects such as Taverna, the Chemistry Development Kit (CDK) or the Waikato Environment for Knowledge Analysis (WEKA). A first integrated version 1.0 of CDK-Taverna was recently released to the public.

RESULTS

The CDK-Taverna project was migrated to the most up-to-date versions of its foundational software libraries with a complete re-engineering of its worker's architecture (version 2.0). 64-bit computing and multi-core usage by paralleled threads are now supported to allow for fast in-memory processing and analysis of large sets of molecules. Earlier deficiencies like workarounds for iterative data reading are removed. The combinatorial chemistry related reaction enumeration features are considerably enhanced. Additional functionality for calculating a natural product likeness score for small molecules is implemented to identify possible drug candidates. Finally the data analysis capabilities are extended with new workers that provide access to the open-source WEKA library for clustering and machine learning as well as training and test set partitioning. The new features are outlined with usage scenarios.

CONCLUSIONS

CDK-Taverna 2.0 as an open-source cheminformatics workflow solution matured to become a freely available and increasingly powerful tool for the biosciences. The combination of the new CDK-Taverna worker family with the already available workflows developed by a lively Taverna community and published on myexperiment.org enables molecular scientists to quickly calculate, process and analyse molecular data as typically found in e.g. today's systems biology scenarios.

CDK-Taverna: an open workflow environment for cheminformatics

Background

Small molecules are of increasing interest for bioinformatics in areas such as metabolomics and drug discovery. The recent release of large open access chemistry databases generates a demand for flexible tools to process them and discover new knowledge. To freely support open science based on these data resources, it is desirable for the processing tools to be open source and available for everyone.

Results

Here we describe a novel combination of the workflow engine Taverna and the cheminformatics library Chemistry Development Kit (CDK) resulting in a open source workflow solution for cheminformatics. We have implemented more than 160 different workers to handle specific cheminformatics tasks. We describe the applications of CDK-Taverna in various usage scenarios.

Conclusions

The combination of the workflow engine Taverna and the Chemistry Development Kit provides the first open source cheminformatics workflow solution for the biosciences. With the Taverna-community working towards a more powerful workflow engine and a more user-friendly user interface, CDK-Taverna has the potential to become a free alternative to existing proprietary workflow tools.