Institut für biologische und chemische Informatik
Refine
Document Type
- Article (22)
- Preprint (11)
- Other (8)
- Conference Proceeding (7)
- Doctoral Thesis (2)
Language
- English (50)
Is part of the Bibliography
- no (50)
Keywords
- Dissipative Particle Dynamics (4)
- Deep Learning (3)
- Transformer (3)
- Artificial Intelligence (2)
- DECIMER (2)
- Deep learning (2)
- OCSR (2)
- OCSR, Optical Chemical Structure Recognition (2)
- AI (1)
- AlphaFold, ColabFold, PyMOL (1)
Institute
Comparative Analysis of Chemical Structure String Representations for Neural Machine Translation
(2025)
In this work, we present a comparative analysis of SMILES, DeepSMILES, and SELFIES string representations for chemical struc-tures in neural machine translation tasks in cheminformatics. Using transformer-based models, we systematically evaluated their effective-ness in translating between these representations and the correspond-ing linguistic IUPAC nomenclature. The experimental results demon-strate comparable performance for all three string representations, with SMILES achieving a marginally higher accuracy (99.30% with stereo-chemical information, 99.21% without) compared to its alternatives. In scaling experiments with 1, 10, and 50 million compounds, the perfor-mance differences remained small, though the performance gap narrowed with larger datasets. These findings suggest that researchers can con-fidently continue using SMILES for neural machine translation tasks with transformers, which benefits from their extensive support in exist-ing chemical libraries, tools, and databases, rather than adopting newer representations. This work has a significant impact on developing more efficient chemical language models in drug discovery, material science, and chemical database curation.
The exponential growth of chemical literature necessitates the development of automated tools for extracting and curating molecular information from unstructured scientific publications into open-access chemical databases. Current optical chemical structure recognition (OCSR) and named entity recognition solutions operate in isolation, which limits their scalability for comprehensive literature curation. Here we present MARCUS (Molecular Annotation and Recognition for Curating Unravelled Structures), a tool to aid curators in performing literature curation in the field of natural products. This integrated web-based platform combines automated text annotation, multi-engine OCSR, and direct submission capabilities to the COCONUT database. MARCUS employs a fine-tuned GPT-4 model to extract chemical entities and utilises an ensemble approach integrating DECIMER, MolNexTR, and MolScribe for structure recognition. The platform aims to streamline the data extraction workflow from PDF upload to database submission, significantly reducing curation time. MARCUS bridges the gap between unstructured chemical literature and machine-actionable databases, enabling FAIR data principles and facilitating AI-driven chemical discovery. Through open-source code, accessible models, and comprehensive documentation, the web application enhances accessibility and promotes community-driven development. This approach facilitates unrestricted use and encourages the collaborative advancement of automated chemical literature curation tools. We dedicate MARCUS to Dr Marcus Ennis, the longest-serving curator of the ChEBI database, on the occasion of his 75th birthday.
PySSA (Python rich client for visual protein Sequence to Structure Analysis) for Windows is a comfortable open Graphical User Interface (GUI) application combining the protein sequence to structure prediction capabilities of ColabFold with the open-source variant of the molecular structure visualization and analysis system PyMOL to make both available to the scientific end-user. PySSA enables the creation and sharing of workflow projects that comprise defined protein 3D structure predictions from their amino acid sequence, protein 3D structure alignments, as well as their visual analysis with distance diagrams or hotspot inspection. All operations can be conveniently performed by scientists without specialized computer skills or even programming knowledge on their local Windows computers, without the need for powerful GPU hardware. Thus, PySSA can help make protein structure prediction more accessible for end-users in scientific research areas like protein chemistry or molecular biology. In addition, the application is well-suited for educational purposes due to its user-friendliness and low learning curve. PySSA is openly available on GitHub, alongside a convenient installer executable for the Windows operating system: https://urban233.github.io/PySSA/install.html. To demonstrate its capabilities, the usage of PySSA in a protein mutation study on the protein drug Bone Morphogenetic Protein 2 (BMP2) is described: the structure prediction results indicate that the previously reported BMP2-2Hep-7M mutant, which is intended to be less prone to aggregation, does not exhibit significant spatial rearrangements of amino acid residues interacting with the receptor.
Naming chemical compounds systematically is a complex task governed by a set of rules established by the International Union of Pure and Applied Chemistry (IUPAC). These rules are universal and widely accepted by chemists worldwide, but their complexity makes it challenging for individuals to consistently apply them accurately. A translation method can be employed to address this challenge. Accurate translation of chemical compounds from SMILES notation into their corresponding IUPAC names is crucial, as it can significantly streamline the laborious process of naming chemical structures. Here, we present STOUT (SMILES-TO-IUPAC-name translator) V2, which addresses this challenge by introducing a transformer-based model that translates string representations of chemical structures into IUPAC names. Trained on a dataset of nearly 1 billion SMILES strings and their corresponding IUPAC names, STOUT V2 demonstrates exceptional accuracy in generating IUPAC names, even for complex chemical structures. The model’s ability to capture intricate patterns and relationships within chemical structures enables it to generate precise and standardised IUPAC names. While established deterministic algorithms remain the gold standard for systematic chemical naming, our work, enabled by access to OpenEye’s Lexichem software through an academic license, demonstrates the potential of neural approaches to complement existing tools in chemical nomenclature.
Naming chemical compounds systematically is a complex task governed by a set of rules established by the International Union of Pure and Applied Chemistry (IUPAC). These rules are universal and widely accepted by chemists worldwide, but their complexity makes it challenging for individuals to consistently apply them accurately. A translation method can be employed to address this challenge. Accurate translation of chemical compounds from SMILES notation into their corresponding IUPAC names is crucial, as it can significantly streamline the laborious process of naming chemical structures. Here, we present STOUT (SMILES-TO-IUPAC-name translator) V2.0, which addresses this challenge by introducing a transformer-based model that translates string representations of chemical structures into IUPAC names. Trained on a dataset of nearly 1 billion SMILES strings and their corresponding IUPAC names, STOUT V2.0 demonstrates exceptional accuracy in generating IUPAC names, even for complex chemical structures. The model's ability to capture intricate patterns and relationships within chemical structures enables it to generate precise and standardised IUPAC names. Deterministic algorithms for systematically naming chemical structures have been available for many years. Also, this work has only been possible through an academic license for OpenEye’s Lexichem software.
An automated pipeline for comprehensive calculation of intermolecular interaction energies based on molecular force-fields using the Tinker molecular modelling package is presented. Starting with non-optimized chemically intuitive monomer structures, the pipeline allows the approximation of global minimum energy monomers and dimers, configuration sampling for various monomer–monomer distances, estimation of coordination numbers by molecular dynamics simulations, and the evaluation of differential pair interaction energies. The latter are used to derive Flory–Huggins parameters and isotropic particle–particle repulsions for Dissipative Particle Dynamics (DPD). The computational results for force fields MM3, MMFF94, OPLS-AA and AMOEBA09 are analyzed with Density Functional Theory (DFT) calculations and DPD simulations for a mixture of the non-ionic polyoxyethylene alkyl ether surfactant C10E4 with water to demonstrate the usefulness of the approach.
From https://github.com/zielesny/MFsim:
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 (DPD). The environment integrates and supports the complete preparation-simulation-evaluation triad of a mesoscopic simulation task. 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 enables customized extensions for different fields of research.
MFsim uses several open libraries (see MFSimVersionHistory.txt for details and references below) and is published as open source under the GNU General Public License version 3 (see LICENSE).
MFsim has been described in the scientific literature and used for DPD studies (see references below).
From https://github.com/zielesny/Jdpd:
Jdpd - An open Java Simulation Kernel for Molecular Fragment Dissipative Particle Dynamics (DPD)
Jdpd is an open Java simulation kernel for Molecular Fragment Dissipative Particle Dynamics (DPD) 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 kernel may be utilized in different simulation environments ranging from flexible scripting solutions up to fully integrated “all-in-one” simulation systems like MFsim.
Since Jdpd version 1.6.1.0 Jdpd is available in a (basic) double-precision version and a (derived) single-precision version (= JdpdSP) for all numerical calculations, where the single precision version needs about half the memory of the double precision version.
Jdpd uses the Apache Commons Math and Apache Commons RNG libraries and is published as open source under the GNU General Public License version 3. This repository comprises the Java bytecode libraries (including the Apache Commons Math and RNG libraries), the Javadoc HTML documentation and the Netbeans source code packages including Unit tests.
Jdpd has been described in the scientific literature (the final manuscript 2018 - van den Broek - Jdpd - Final Manucsript.pdf is added to the repository) and used for DPD studies (see references below).
See text file JdpdVersionHistory.txt for a version history with more detailed information.
Computational methods for the accurate prediction of protein folding based on amino acid sequences have been researched for decades. The field has been significantly advanced in recent years by deep learning-based approaches, like AlphaFold, RoseTTAFold, or ColabFold. Although these can be used by the scientific community in various, mostly free and open ways, they are not yet widely used by bench scientists in relevant fields such as protein biochemistry or molecular biology, who are often not familiar with software tools such as scripting notebooks, command-line interfaces or cloud computing. In addition, visual inspection functionalities like protein structure displays, structure alignments, and specific protein hotspot analyses are required as a second step to interpret and apply the predicted structures in ongoing research studies.
PySSA (Python rich client for visual protein Sequence to Structure Analysis) is an open Graphical User Interface (GUI) application combining the protein sequence to structure prediction capabilities of ColabFold with the open-source variant of the molecular structure visualisation and analysis system PyMOL to make both available to the scientific end-user. PySSA enables the creation of managed and shareable projects with defined protein structure prediction and corresponding alignment workflows that can be conveniently performed by scientists without specialised computer skills or programming knowledge on their local computers. Thus, PySSA can help make protein structure prediction more accessible for end-users in protein chemistry and molecular biology as well as be used for educational purposes. It is openly available on GitHub, alongside a custom graphical installer executable for the Windows operating system: https://github.com/urban233/PySSA/wiki/Installation-for-Windows-Operating-System.
To demonstrate the capabilities of PySSA, its usage in a protein mutation study on the protein drug Bone Morphogenetic Protein 2 (BMP2) is described: the structure prediction results indicate that the previously reported BMP2-2Hep-7M mutant, which is intended to be less prone to aggregation, does not exhibit significant spatial rearrangements of amino acid residues interacting with the receptor.
An automated pipeline for comprehensive calculation of intermolecular interaction energies based on molecular force-fields using the Tinker molecular modelling package is presented. Starting with non-optimized chemically intuitive monomer structures, the pipeline allows the approximation of global minimum energy monomers and dimers, configuration sampling for various monomer-monomer distances, estimation of coordination numbers by molecular dynamics simulations, and the evaluation of differential pair interaction energies. The latter are used to derive Flory-Huggins parameters and isotropic particle-particle repulsions for Dissipative Particle Dynamics (DPD). The computational results for force fields MM3, MMFF94, OPLS-AA and AMOEBA09 are analyzed with Density Functional Theory (DFT) calculations and DPD simulations for a mixture of the non-ionic polyoxyethylene alkyl ether surfactant C10E4 with water to demonstrate the usefulness of the approach.

