Structure-based classification and ontology in chemistry

Mixtures Recomposition by Neural Nets: A Multidisciplinary Overview

Artificial Neural Networks (ANNs) are transforming how we understand chemical mixtures, providing an expressive view of the chemical space and multiscale processes. Their hybridization with physical knowledge can bridge the gap between predictivity and understanding of the underlying processes. This overview explores recent progress in ANNs, particularly their potential in the 'recomposition' of chemical mixtures. Graph-based representations reveal patterns among mixture components, and deep learning models excel in capturing complexity and symmetries when compared to traditional Quantitative Structure-Property Relationship models. Key components, such as Hamiltonian networks and convolution operations, play a central role in representing multiscale mixtures. The integration of ANNs with Chemical Reaction Networks and Physics-Informed Neural Networks for inverse chemical kinetic problems is also examined. The combination of sensors with ANNs shows promise in optical and biomimetic applications. A common ground is identified in the context of statistical physics, where ANN-based methods iteratively adapt their models by blending their initial states with training data. The concept of mixture recomposition unveils a reciprocal inspiration between ANNs and reactive mixtures, highlighting learning behaviors influenced by the training environment.

Ontologies4Chem: the landscape of ontologies in chemistry

For a long time, databases such as CAS, Reaxys, PubChem or ChemSpider mostly rely on unique numerical identifiers or chemical structure identifiers like InChI, SMILES or others to link data across heterogeneous data sources. The retrospective processing of information and fragmented data from text publications to maintain these databases is a cumbersome process. Ontologies are a holistic approach to semantically describe data, information and knowledge of a domain. They provide terms, relations and logic to semantically annotate and link data building knowledge graphs. The application of standard taxonomies and vocabularies from the very beginning of data generation and along research workflows in electronic lab notebooks (ELNs), software tools, and their final publication in data repositories create FAIR data straightforwardly. Thus a proper semantic description of an investigation and the why, how, where, when, and by whom data was produced in conjunction with the description and representation of research data is a natural outcome in contrast to the retrospective processing of research publications as we know it. In this work we provide an overview of ontologies in chemistry suitable to represent concepts of research and research data. These ontologies are evaluated against several criteria derived from the FAIR data principles and their possible application in the digitisation of research data management workflows.

How different are marine microbial natural products compared to their terrestrial counterparts?

Covering: 1877 to 2020A key challenge in natural products research is the selection of biodiversity to yield novel chemistry. Recently, marine microorganisms have become a preferred source. But how novel are marine microorganism natural products compared to those reported from terrestrial microbes? Cluster analysis of chemical fingerprints and molecular scaffold analysis of 55 817 compounds reported from marine and terrestrial microorganisms, and marine macro-organisms showed that 76.7% of the compounds isolated from marine microorganisms are closely related to compounds isolated from terrestrial microorganisms. Only 14.3% of marine microorganism natural products are unique when marine macro-organism natural products are also considered. Studies targeting marine specific and understudied microbial phyla result in a higher likelihood of finding marine specific compounds, whereas the depth and geographic location of microorganism collection have little influence. We recommend marine targeted strain isolation, incorporating early use of genomic sequencing to guide strain selection, innovation in culture media and cultivation techniques and the application of cheminformatics tools to focus on unique natural product diversity, rather than the dereplication of known compounds.

An immunologically friendly classification of non-peptidic ligands

The Immune Epitope Database (IEDB) freely provides experimental data regarding immune epitopes to the scientific public. The main users of the IEDB are immunologists who can easily use our web interface to search for peptidic epitopes via their simple single-letter codes. For example, 'A' stands for 'alanine'. Similarly, users can easily navigate the IEDB's simplified NCBI taxonomy hierarchy to locate proteins from specific organisms. However, some epitopes are non-peptidic, such as carbohydrates, lipids, chemicals and drugs, and it is more challenging to consistently name them and search upon, making access to their data more problematic for immunologists. Therefore, we set out to improve access to non-peptidic epitope data in the IEDB through the simplification of the non-peptidic hierarchy used in our search interfaces. Here, we present these efforts and their outcomes. Database URL:  http://www.iedb.org/.

Learning chemistry: exploring the suitability of machine learning for the task of structure-based chemical ontology classification

Chemical data is increasingly openly available in databases such as PubChem, which contains approximately 110 million compound entries as of February 2021. With the availability of data at such scale, the burden has shifted to organisation, analysis and interpretation. Chemical ontologies provide structured classifications of chemical entities that can be used for navigation and filtering of the large chemical space. ChEBI is a prominent example of a chemical ontology, widely used in life science contexts. However, ChEBI is manually maintained and as such cannot easily scale to the full scope of public chemical data. There is a need for tools that are able to automatically classify chemical data into chemical ontologies, which can be framed as a hierarchical multi-class classification problem. In this paper we evaluate machine learning approaches for this task, comparing different learning frameworks including logistic regression, decision trees and long short-term memory artificial neural networks, and different encoding approaches for the chemical structures, including cheminformatics fingerprints and character-based encoding from chemical line notation representations. We find that classical learning approaches such as logistic regression perform well with sets of relatively specific, disjoint chemical classes, while the neural network is able to handle larger sets of overlapping classes but needs more examples per class to learn from, and is not able to make a class prediction for every molecule. Future work will explore hybrid and ensemble approaches, as well as alternative network architectures including neuro-symbolic approaches.

Knowledge Graph Approach to Combustion Chemistry and Interoperability

In this paper, we demonstrate through examples how the concept of a Semantic Web based knowledge graph can be used to integrate combustion modeling into cross-disciplinary applications and in particular how inconsistency issues in chemical mechanisms can be addressed. We discuss the advantages of linked data that form the essence of a knowledge graph and how we implement this in a number of interconnected ontologies, specifically in the context of combustion chemistry. Central to this is OntoKin, an ontology we have developed for capturing both the content and the semantics of chemical kinetic reaction mechanisms. OntoKin is used to represent the example mechanisms from the literature in a knowledge graph, which itself is part of the existing, more general knowledge graph and ecosystem of autonomous software agents that are acting on it. We describe a web interface, which allows users to interact with the system, upload and compare the existing mechanisms, and query species and reactions across the knowledge graph. The utility of the knowledge-graph approach is demonstrated for two use-cases: querying across multiple mechanisms from the literature and modeling the atmospheric dispersion of pollutants emitted by ships. As part of the query use-case, our ontological tools are applied to identify variations in the rate of a hydrogen abstraction reaction from methane as represented by 10 different mechanisms.

Co-exposure to environmental endocrine disruptors in the US population

Exposure to environmental endocrine disruptors (EEDs) has been linked to adverse health outcomes. The vast majority of studies examined one class of EEDs at a time but humans often are exposed to multiple EEDs at the same time. It is, therefore, important to know the co-exposure status of multiple EEDs in an individual, to preclude and control for potential confounding effects posed by co-exposed EEDs. This study examined the concentrations of seven classes of EEDs in the US population utilizing the data from the National Health and Nutrition Examination Survey (NHANES), 2009-2014 survey cycles. We applied linear correlation and cluster analysis to characterize the correlation profile and cluster patterns of these EEDs. We found that EEDs with a similar structure are often highly correlated. Among between-class correlations, mercury and perfluoroalkyl substances (PFAS) and cadmium and polycyclic aromatic hydrocarbons (PAHs) were two significantly correlated EEDs. In epidemiologic studies, measurement and control for co-exposure to pollutants, especially those with similar biological effects, are critical when attempting to make causal inferences. Appropriate statistical methods to handle within- and between-class correlations are needed.

PubChem chemical structure standardization

BACKGROUND

PubChem is a chemical information repository, consisting of three primary databases: Substance, Compound, and BioAssay. When individual data contributors submit chemical substance descriptions to Substance, the unique chemical structures are extracted and stored into Compound through an automated process called structure standardization. The present study describes the PubChem standardization approaches and analyzes them for their success rates, reasons that cause structures to be rejected, and modifications applied to structures during the standardization process. Furthermore, the PubChem standardization is compared to the structure normalization of the IUPAC International Chemical Identifier (InChI) software, as manifested by conversion of the InChI back into a chemical structure.

RESULTS

The observed rejection rate for substances processed by PubChem standardization was 0.36%, which is predominantly attributed to structures with invalid atom valences that cannot be readily corrected without additional information from contributors. Of all structures that pass standardization, 44% are modified in the process, reducing the count of unique structures from 53,574,724 in substance to 45,808,881 in compound as identified by de-aromatized canonical isomeric SMILES. Even though the processing time is very low on average (only 0.4% of structures have individual standardization time above 0.1 s), total standardization time is completely dominated by edge cases: 90% of the time to standardize all structures in PubChem substance is spent on the 2.05% of structures with the highest individual standardization time. It is worth noting that 60% of the structures obtained from PubChem structure standardization are not identical to the chemical structure resulting from the InChI (primarily due to preferences for a different tautomeric form).

CONCLUSIONS

Standardization of chemical structures is complicated by the diversity of chemical information and their representations approaches. The PubChem standardization is an effective and efficient tool to account for molecular diversity and to eliminate invalid/incomplete structures. Further development will concentrate on improved tautomer consideration and an expanded stereocenter definition. Modifications are difficult to thoroughly validate, with slight changes often affecting many thousands of structures and various edge cases. The PubChem structure standardization service is accessible as a public resource ( https://pubchem.ncbi.nlm.nih.gov/standardize ), and via programmatic interfaces.

Advances in computational metabolomics and databases deepen the understanding of metabolisms

Mass spectrometry (MS)-based metabolomics is the popular platform for metabolome analyses. Computational techniques for the processing of MS raw data, for example, feature detection, peak alignment, and the exclusion of false-positive peaks, have been established. The next stage of untargeted metabolomics would be to decipher the mass fragmentation of small molecules for the global identification of human-, animal-, plant-, and microbiota metabolomes, resulting in a deeper understanding of metabolisms. This review is an update on the latest computational metabolomics including known/expected structure databases, chemical ontology classifications, and mass spectrometry cheminformatics for the interpretation of mass fragmentations and for the elucidation of unknown metabolites. The importance of metabolome 'databases' and 'repositories' is also discussed because novel biological discoveries are often attributable to the accumulation of data, to relational databases, and to their statistics. Lastly, a practical guide for metabolite annotations is presented as the summary of this review.

Encoding of coordination complexes with XML

An in-silico system to encode structure, bonding and properties of coordination complexes is developed. The encoding is achieved through a semantic XML markup frame. Composition of the coordination complexes is captured in terms of central atom and ligands. Structural information of central atom is detailed in terms of electron status of valence electron orbitals. The ligands are encoded with specific reference to the electron environment of ligand centre atoms. Behaviour of ligands to form low or high spin complexes is accomplished by assigning a Ligand Centre Value to every ligand based on the electronic environment of ligand centre atom. Chemical ontologies are used for categorization purpose and to control different hybridization schemes. Complexes formed by the central atoms of transition metal, non-transition elements belonging to s-block, p-block and f-block are encoded with a generic encoding platform. Complexes of homoleptic, heteroleptic and bridged types are also covered by this encoding system. Utility of the encoded system to predict redox electron transfer reaction in the coordination complexes is demonstrated with a simple application.

ClassyFire: automated chemical classification with a comprehensive, computable taxonomy

Background

Scientists have long been driven by the desire to describe, organize, classify, and compare objects using taxonomies and/or ontologies. In contrast to biology, geology, and many other scientific disciplines, the world of chemistry still lacks a standardized chemical ontology or taxonomy. Several attempts at chemical classification have been made; but they have mostly been limited to either manual, or semi-automated proof-of-principle applications. This is regrettable as comprehensive chemical classification and description tools could not only improve our understanding of chemistry but also improve the linkage between chemistry and many other fields. For instance, the chemical classification of a compound could help predict its metabolic fate in humans, its druggability or potential hazards associated with it, among others. However, the sheer number (tens of millions of compounds) and complexity of chemical structures is such that any manual classification effort would prove to be near impossible.

Results

We have developed a comprehensive, flexible, and computable, purely structure-based chemical taxonomy (ChemOnt), along with a computer program (ClassyFire) that uses only chemical structures and structural features to automatically assign all known chemical compounds to a taxonomy consisting of >4800 different categories. This new chemical taxonomy consists of up to 11 different levels (Kingdom, SuperClass, Class, SubClass, etc.) with each of the categories defined by unambiguous, computable structural rules. Furthermore each category is named using a consensus-based nomenclature and described (in English) based on the characteristic common structural properties of the compounds it contains. The ClassyFire webserver is freely accessible at http://classyfire.wishartlab.com/. Moreover, a Ruby API version is available at https://bitbucket.org/wishartlab/classyfire_api, which provides programmatic access to the ClassyFire server and database. ClassyFire has been used to annotate over 77 million compounds and has already been integrated into other software packages to automatically generate textual descriptions for, and/or infer biological properties of over 100,000 compounds. Additional examples and applications are provided in this paper.

Conclusion

ClassyFire, in combination with ChemOnt (ClassyFire’s comprehensive chemical taxonomy), now allows chemists and cheminformaticians to perform large-scale, rapid and automated chemical classification. Moreover, a freely accessible API allows easy access to more than 77 million “ClassyFire” classified compounds. The results can be used to help annotate well studied, as well as lesser-known compounds. In addition, these chemical classifications can be used as input for data integration, and many other cheminformatics-related tasks.

Electronic supplementary material

The online version of this article (doi:10.1186/s13321-016-0174-y) contains supplementary material, which is available to authorized users.

Harnessing ontology and machine learning for RSO classification

Classification is an important part of resident space objects (RSOs) identification, which is a main focus of space situational awareness. Owing to the absence of some features caused by the limited and uncertain observations, RSO classification remains a difficult task. In this paper, an ontology for RSO classification named OntoStar is built upon domain knowledge and machine learning rules. Then data describing RSO are represented by OntoStar. A demo shows how an RSO is classified based on OntoStar. It is also shown in the demo that traceable and comprehensible reasons for the classification can be given, hence the classification can be checked and validated. Experiments on WEKA show that ontology-based classification gains a relatively high accuracy and precision for classifying RSOs. When classifying RSOs with imperfect data, ontology-based classification keeps its performances, showing evident advantages over classical machine learning classifiers who either have increases of 5 % at least in FP rate or have decreases of 5 % at least in indexes such as accuracy, precision and recall.

libChEBI: an API for accessing the ChEBI database

Background

ChEBI is a database and ontology of chemical entities of biological interest. It is widely used as a source of identifiers to facilitate unambiguous reference to chemical entities within biological models, databases, ontologies and literature. ChEBI contains a wealth of chemical data, covering over 46,500 distinct chemical entities, and related data such as chemical formula, charge, molecular mass, structure, synonyms and links to external databases. Furthermore, ChEBI is an ontology, and thus provides meaningful links between chemical entities. Unlike many other resources, ChEBI is fully human-curated, providing a reliable, non-redundant collection of chemical entities and related data. While ChEBI is supported by a web service for programmatic access and a number of download files, it does not have an API library to facilitate the use of ChEBI and its data in cheminformatics software.

Results

To provide this missing functionality, libChEBI, a comprehensive API library for accessing ChEBI data, is introduced. libChEBI is available in Java, Python and MATLAB versions from http://github.com/libChEBI, and provides full programmatic access to all data held within the ChEBI database through a simple and documented API. libChEBI is reliant upon the (automated) download and regular update of flat files that are held locally. As such, libChEBI can be embedded in both on- and off-line software applications.

Conclusions

libChEBI allows better support of ChEBI and its data in the development of new cheminformatics software. Covering three key programming languages, it allows for the entirety of the ChEBI database to be accessed easily and quickly through a simple API. All code is open access and freely available.

Electronic supplementary material

The online version of this article (doi:10.1186/s13321-016-0123-9) contains supplementary material, which is available to authorized users.

The BioDICE Taverna plugin for clustering and visualization of biological data: a workflow for molecular compounds exploration

Background

In many experimental pipelines, clustering of multidimensional biological datasets is used to detect hidden structures in unlabelled input data. Taverna is a popular workflow management system that is used to design and execute scientific workflows and aid in silico experimentation. The availability of fast unsupervised methods for clustering and visualization in the Taverna platform is important to support a data-driven scientific discovery in complex and explorative bioinformatics applications.

Results

This work presents a Taverna plugin, the Biological Data Interactive Clustering Explorer (BioDICE), that performs clustering of high-dimensional biological data and provides a nonlinear, topology preserving projection for the visualization of the input data and their similarities. The core algorithm in the BioDICE plugin is Fast Learning Self Organizing Map (FLSOM), which is an improved variant of the Self Organizing Map (SOM) algorithm. The plugin generates an interactive 2D map that allows the visual exploration of multidimensional data and the identification of groups of similar objects. The effectiveness of the plugin is demonstrated on a case study related to chemical compounds.

Conclusions

The number and variety of available tools and its extensibility have made Taverna a popular choice for the development of scientific data workflows. This work presents a novel plugin, BioDICE, which adds a data-driven knowledge discovery component to Taverna. BioDICE provides an effective and powerful clustering tool, which can be adopted for the explorative analysis of biological datasets.

A rule-based ontological framework for the classification of molecules

BACKGROUND

A variety of key activities within life sciences research involves integrating and intelligently managing large amounts of biochemical information. Semantic technologies provide an intuitive way to organise and sift through these rapidly growing datasets via the design and maintenance of ontology-supported knowledge bases. To this end, OWL-a W3C standard declarative language- has been extensively used in the deployment of biochemical ontologies that can be conveniently organised using the classification facilities of OWL-based tools. One of the most established ontologies for the chemical domain is ChEBI, an open-access dictionary of molecular entities that supplies high quality annotation and taxonomical information for biologically relevant compounds. However, ChEBI is being manually expanded which hinders its potential to grow due to the limited availability of human resources.

RESULTS

In this work, we describe a prototype that performs automatic classification of chemical compounds. The software we present implements a sound and complete reasoning procedure of a formalism that extends datalog and builds upon an off-the-shelf deductive database system. We capture a wide range of chemical classes that are not expressible with OWL-based formalisms such as cyclic molecules, saturated molecules and alkanes. Furthermore, we describe a surface 'less-logician-like' syntax that allows application experts to create ontological descriptions of complex biochemical objects without prior knowledge of logic. In terms of performance, a noticeable improvement is observed in comparison with previous approaches. Our evaluation has discovered subsumptions that are missing from the manually curated ChEBI ontology as well as discrepancies with respect to existing subclass relations. We illustrate thus the potential of an ontology language suitable for the life sciences domain that exhibits a favourable balance between expressive power and practical feasibility.

CONCLUSIONS

Our proposed methodology can form the basis of an ontology-mediated application to assist biocurators in the production of complete and error-free taxonomies. Moreover, such a tool could contribute to a more rapid development of the ChEBI ontology and to the efforts of the ChEBI team to make annotated chemical datasets available to the public. From a modelling point of view, our approach could stimulate the adoption of a different and expressive reasoning paradigm based on rules for which state-of-the-art and highly optimised reasoners are available; it could thus pave the way for the representation of a broader spectrum of life sciences and biomedical knowledge.

Exploiting disjointness axioms to improve semantic similarity measures

MOTIVATION

Representing domain knowledge in biology has traditionally been accomplished by creating simple hierarchies of classes with textual annotations. Recently, expressive ontology languages, such as Web Ontology Language, have become more widely adopted, supporting axioms that express logical relationships other than class-subclass, e.g. disjointness. This is improving the coverage and validity of the knowledge contained in biological ontologies. However, current semantic tools still need to adapt to this more expressive information. In this article, we propose a method to integrate disjointness axioms, which are being incorporated in real-world ontologies, such as the Gene Ontology and the chemical entities of biological interest ontology, into semantic similarity, the measure that estimates the closeness in meaning between classes.

RESULTS

We present a modification of the measure of shared information content, which extends the base measure to allow the incorporation of disjointness information. To evaluate our approach, we applied it to several randomly selected datasets extracted from the chemical entities of biological interest ontology. In 93.8% of these datasets, our measure performed better than the base measure of shared information content. This supports the idea that semantic similarity is more accurate if it extends beyond the hierarchy of classes of the ontology.

CONTACT

joao.ferreira@lasige.di.fc.ul.pt.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

A conceptual basis to encode and detect organic functional groups in XML

A conceptual basis to define and detect organic functional groups is developed. The basic model of a functional group is termed as a primary functional group and is characterized by a group center composed of one or more group center atoms bonded to terminal atoms and skeletal carbon atoms. The generic group center patterns are identified from the structures of known functional groups. Accordingly, a chemical ontology 'Font' is developed to organize the existing functional groups as well as the new ones to be defined by the chemists. The basic model is extended to accommodate various combinations of primary functional groups as functional group assemblies. A concept of skeletal group is proposed to define the characteristic groups composed of only carbon atoms to be regarded as equivalent to functional groups. The combination of primary functional groups with skeletal groups is categorized as skeletal group assembly. In order to make the model suitable for reaction modeling purpose, a Graphical User Interface (GUI) is developed to define the functional groups and to encode in XML format appropriate to detect them in chemical structures. The system is capable of detecting multiple instances of primary functional groups as well as the overlapping poly-functional groups as the respective assemblies.

Automated compound classification using a chemical ontology

Background

Classification of chemical compounds into compound classes by using structure derived descriptors is a well-established method to aid the evaluation and abstraction of compound properties in chemical compound databases. MeSH and recently ChEBI are examples of chemical ontologies that provide a hierarchical classification of compounds into general compound classes of biological interest based on their structural as well as property or use features. In these ontologies, compounds have been assigned manually to their respective classes. However, with the ever increasing possibilities to extract new compounds from text documents using name-to-structure tools and considering the large number of compounds deposited in databases, automated and comprehensive chemical classification methods are needed to avoid the error prone and time consuming manual classification of compounds.

Results

In the present work we implement principles and methods to construct a chemical ontology of classes that shall support the automated, high-quality compound classification in chemical databases or text documents. While SMARTS expressions have already been used to define chemical structure class concepts, in the present work we have extended the expressive power of such class definitions by expanding their structure-based reasoning logic. Thus, to achieve the required precision and granularity of chemical class definitions, sets of SMARTS class definitions are connected by OR and NOT logical operators. In addition, AND logic has been implemented to allow the concomitant use of flexible atom lists and stereochemistry definitions. The resulting chemical ontology is a multi-hierarchical taxonomy of concept nodes connected by directed, transitive relationships.

Conclusions

A proposal for a rule based definition of chemical classes has been made that allows to define chemical compound classes more precisely than before. The proposed structure-based reasoning logic allows to translate chemistry expert knowledge into a computer interpretable form, preventing erroneous compound assignments and allowing automatic compound classification. The automated assignment of compounds in databases, compound structure files or text documents to their related ontology classes is possible through the integration with a chemical structure search engine. As an application example, the annotation of chemical structure files with a prototypic ontology is demonstrated.