Beyond symmetric self-assembly and effective molarity: unlocking functional enzyme mimics with robust organic cages

• Keith G. Andrews

Beilstein J. Org. Chem. 2025, 21, 421–443, doi:10.3762/bjoc.21.30

• enabled by improving automated experimental screening of existing cavities for new activity (sensing, catalysis) [418]. Crucially, improved access to experimental structure–activity relationships of incrementally developed cavities [21] is required to feed rational or machine learning advances. The unique

Emerging trends in the optimization of organic synthesis through high-throughput tools and machine learning

• Pablo Quijano Velasco,
• Kedar Hippalgaonkar and
• Balamurugan Ramalingam

Beilstein J. Org. Chem. 2025, 21, 10–38, doi:10.3762/bjoc.21.3

• chemical reaction optimization has been enabled by advances in lab automation and the introduction of machine learning algorithms. Therein, multiple reaction variables can be synchronously optimized to obtain the optimal reaction conditions, requiring a shorter experimentation time and minimal human

• intervention. Herein, we review the currently used state-of-the-art high-throughput automated chemical reaction platforms and machine learning algorithms that drive the optimization of chemical reactions, highlighting the limitations and future opportunities of this new field of research. Keywords: autonomous

• reactors; data processing; high-throughput experimentation; machine learning; reaction optimization; Introduction Organic synthesis plays a crucial role in drug discovery, polymer synthesis, materials science, agrochemicals, and specialty chemicals. Their synthesis and process optimization require

Chemical structure metagenomics of microbial natural products: surveying nonribosomal peptides and beyond

• Thomas Ma and
• John Chu

Beilstein J. Org. Chem. 2024, 20, 3050–3060, doi:10.3762/bjoc.20.253

• contrast, despite being one of the most common BBs in NRPs, no ornithine is β-hydroxylated to the best of our knowledge. Last but not least, natural product research has forayed into the use of artificial intelligence (AI) tools. For example, Magarvey and co-workers used machine learning to help improve

Machine learning-guided strategies for reaction conditions design and optimization

• Lung-Yi Chen and
• Yi-Pei Li

Beilstein J. Org. Chem. 2024, 20, 2476–2492, doi:10.3762/bjoc.20.212

• /bjoc.20.212 Abstract This review surveys the recent advances and challenges in predicting and optimizing reaction conditions using machine learning techniques. The paper emphasizes the importance of acquiring and processing large and diverse datasets of chemical reactions, and the use of both global

• efficiency and effectiveness of reaction conditions design, and enable novel discoveries in synthetic chemistry. Keywords: data preprocessing; reaction conditions prediction; reaction data mining; reaction optimization; reaction representation; Introduction Machine learning (ML) techniques have been widely

• machine-learning solutions for reaction diagram parsing [76][77], there are still some limitations. For instance, sometimes the reaction conditions are listed in tables, and certain functional groups in images are represented by abbreviations (e.g., R-groups). To achieve more complete data extraction

Catalysing (organo-)catalysis: Trends in the application of machine learning to enantioselective organocatalysis

• Stefan P. Schmid,
• Leon Schlosser,
• Frank Glorius and
• Kjell Jorner

Beilstein J. Org. Chem. 2024, 20, 2280–2304, doi:10.3762/bjoc.20.196

• interest over the last decades. Concurrent to this development, machine learning (ML) has been increasingly applied in the chemical domain to efficiently uncover hidden patterns in data and accelerate scientific discovery. While the uptake of ML in organocatalysis has been comparably slow, the last two

• . Keywords: catalyst design; machine learning; modelling; organocatalysis; selectivity prediction; Introduction Since the beginning of the 21st century, organocatalysts [1] have established themselves as a third group of homogeneous catalysts, next to biocatalysts [2] (enzymes) and transition metal-based

• between catalyst, substrate and the DABCOnium moiety. Subsequently, a random forest model was used to predict exo/endo- and regioselectivity of the reaction. Using random forest as an interpretable machine learning model allowed to extract the important features of the model, which indicated that the

Finding the most potent compounds using active learning on molecular pairs

• Zachary Fralish and
• Daniel Reker

Beilstein J. Org. Chem. 2024, 20, 2152–2162, doi:10.3762/bjoc.20.185

• regimes by enabling faster and more accurate identification of more diverse molecular hits against critical drug targets. Keywords: active learning; drug design; machine learning; molecular optimization; potency predictions; Introduction Active learning is a powerful concept in molecular machine

• insufficient to support the accurate training of data-hungry machine learning models [10][11] and thereby leading to potentially sub-optimal experimental design due to an incomplete understanding of the underlying structure–activity relationship and poor calibration of predictive uncertainty. Additionally

• property differences between molecules [12]. Compared to classic molecular machine learning algorithms, which are trained to predict absolute property values, such paired approaches are more well-equipped to guide molecular optimization by directly learning from and predicting molecular property

Computational toolbox for the analysis of protein–glycan interactions

• Ferran Nieto-Fabregat,
• Maria Pia Lenza,
• Angela Marseglia,
• Cristina Di Carluccio,
• Antonio Molinaro,
• Alba Silipo and
• Roberta Marchetti

Beilstein J. Org. Chem. 2024, 20, 2084–2107, doi:10.3762/bjoc.20.180

• design and discovery providing access to physics-based molecular modelling tools and machine learning technologies from a single modelling environment, however, it is not free-accessible. Further valuable insights into the structure and conformation of saccharides, determined by experiment and simulation

• its amino acid sequence and the structure of a related protein that is already known. However, also template-free PSP has obtained significant progress recently via machine learning and search-based optimisation approaches [87]. There are several software programs and tools available for homology

• modelling, and some of the most popular include: 1. AlphaFold2 [88]: It is an open-access protein structure prediction system based on artificial intelligence and machine learning. It is based on a neural network that can predict the 3D protein structure at a high accuracy level. The AlphaFold solution is

Hetero-polycyclic aromatic systems: A data-driven investigation of structure–property relationships

• Sabyasachi Chakraborty,
• Eduardo Mayo Yanes and
• Renana Gershoni-Poranne

Beilstein J. Org. Chem. 2024, 20, 1817–1830, doi:10.3762/bjoc.20.160

• relationships require more advanced data-analysis tools, and we are currently leveraging different machine learning and deep learning techniques to tap the full potential of the COMPAS-2 dataset. A) The building blocks used in the COMPAS-2 datasets. B) Possible annulation types formed when combining the

Discovery of antimicrobial peptides clostrisin and cellulosin from Clostridium: insights into their structures, co-localized biosynthetic gene clusters, and antibiotic activity

• Moisés Alejandro Alejo Hernandez,
• Katia Pamela Villavicencio Sánchez,
• Rosendo Sánchez Morales,
• Karla Georgina Hernández-Magro Gil,
• David Silverio Moreno-Gutiérrez,
• Eddie Guillermo Sanchez-Rueda,
• Yanet Teresa-Cruz,
• Brian Choi,
• Armando Hernández Garcia,
• Alba Romero-Rodríguez,
• Oscar Juárez,
• Siseth Martínez-Caballero,
• Mario Figueroa and
• Corina-Diana Ceapă

Beilstein J. Org. Chem. 2024, 20, 1800–1816, doi:10.3762/bjoc.20.159

• compounds. For example, the discovery of teixobactin [8] has paved the way for developing antibiotics with innovative mechanisms of action. Similarly, the discovery of halicin, a molecule identified through a machine learning-based approach, has shown promise as a broad-spectrum antimicrobial agent

pKalculator: A pK_a predictor for C–H bonds

• Rasmus M. Borup,
• Nicolai Ree and
• Jan H. Jensen

Beilstein J. Org. Chem. 2024, 20, 1614–1622, doi:10.3762/bjoc.20.144

• . As molecular complexity increases, this task becomes more challenging. This paper introduces pKalculator, a quantum chemistry (QM)-based workflow for automatic computations of C–H pKa values, which is used to generate a training dataset for a machine learning (ML) model. The QM workflow is

• (DMSO) using a graph convolutional neural network (GCNN) [3]. Using a mix of experimental and computed pKa data, they achieved a mean absolute error (MAE) of 2.1 pKa units. Lee and co-workers also addressed this problem by creating a general machine learning (ML) model using either a neural network or

• values for all deprotonated C–H sites using Equation 2: Machine learning The feature descriptor Recent research shows that the atomic descriptors introduced by Finkelmann et al. [30][31], using charge model 5 (CM5) atomic charges [32], are a great representation of atoms in molecules that can be used in

Generation of multimillion chemical space based on the parallel Groebke–Blackburn–Bienaymé reaction

• Evgen V. Govor,
• Vasyl Naumchyk,
• Ihor Nestorak,
• Dmytro S. Radchenko,
• Dmytro Dudenko,
• Yurii S. Moroz,
• Olexiy D. Kachkovsky and
• Oleksandr O. Grygorenko

Beilstein J. Org. Chem. 2024, 20, 1604–1613, doi:10.3762/bjoc.20.143

• as high-throughput docking or machine learning [33][34][35][36][37]. In this work, we aimed at the implementation of the GBB reaction for the generation of such ultra-large chemical space, including experimental evaluation of the synthesis success rate (SSR, i.e., percentage of experiments that

Predicting bond dissociation energies of cyclic hypervalent halogen reagents using DFT calculations and graph attention network model

• Yingbo Shao,
• Zhiyuan Ren,
• Zhihui Han,
• Li Chen,
• Yao Li and
• Xiao-Song Xue

Beilstein J. Org. Chem. 2024, 20, 1444–1452, doi:10.3762/bjoc.20.127

• homolytic BDEs across different halogen centers, while a strong linear correlation was noted among the heterolytic BDEs across these centers. Furthermore, we developed a predictive model for both homolytic and heterolytic BDEs of cyclic hypervalent halogen compounds using machine learning algorithms. The

• ; machine learning; Introduction Hypervalent iodine reagents are increasingly gaining attention in the fields of organic synthesis and catalysis due to their environmental benefits, accessibility, and cost-efficiency [1][2][3][4][5][6][7][8][9][10][11]. Over the last three decades, a series of cyclic

• especially crucial for designing novel reagents. Yet, the BDE values of hypervalent bromine(III) and chlorine(III) reagents remain largely elusive, hampering the design and synthesis of novel reagents. In recent years, machine learning has emerged as a promising and cost-effective alternative to traditional

GlAIcomics: a deep neural network classifier for spectroscopy-augmented mass spectrometric glycans data

• Thomas Barillot,
• Baptiste Schindler,
• Baptiste Moge,
• Elisa Fadda,
• Franck Lépine and
• Isabelle Compagnon

Beilstein J. Org. Chem. 2023, 19, 1825–1831, doi:10.3762/bjoc.19.134

• analysis, thus excluding manual interpretation. Besides, in the prospective of deploying the technology beyond the molecular spectroscopy community, it is essential to develop an automated, reliable, and robust strategy for the analysis of the spectroscopic data. Machine learning methods appear to be

• spectra for cancer classification [8] and many research groups focused their efforts on using machine learning for simulating molecular structures; generating vibrational spectra; and classifying chemical groups based on vibrational features [9][10]. In a recent publication, the random forest approach was

• proposed to identify the presence of structural features in oligosaccharides based on their gas-phase IR spectra [11]. To the best of our knowledge, machine learning classification studies have not been reported to identify saccharides using MS–IR carbohydrate analysis. Here, we report a study of a

Functional characterisation of twelve terpene synthases from actinobacteria

• Anuj K. Chhalodia,
• Houchao Xu,
• Georges B. Tabekoueng,
• Binbin Gu,
• Kizerbo A. Taizoumbe,
• Lukas Lauterbach and
• Jeroen S. Dickschat

Beilstein J. Org. Chem. 2023, 19, 1386–1398, doi:10.3762/bjoc.19.100

• for a pathway reconstruction towards artemisinin. The increased knowledge about terpene synthases together with the structures of their products will also be of interest for machine learning approaches to enable the prediction of terpene synthase functions from their amino acid sequences. Both aspects

Bromination of endo-7-norbornene derivatives revisited: failure of a computational NMR method in elucidating the configuration of an organic structure

• Demet Demirci Gültekin,
• Arif Daştan,
• Yavuz Taşkesenligil,
• Cavit Kazaz,
• Yunus Zorlu and
• Metin Balci

Beilstein J. Org. Chem. 2023, 19, 764–770, doi:10.3762/bjoc.19.56

• Kutateladze claimed that based on an applied machine learning-augmented DFT method for computational NMR that the structure of the product, (1R,2R,3S,4S,7s)-2,3,7-tribromobicyclo[2.2.1]heptane was wrong. With the aid of their computational method, they revised a number of published structures, including ours

• have developed a machine learning-augmented DFT method for computational NMR, DU8ML, for fast and ‘accurate’ computational approaches [2]. They applied this computational method to a number of previously published organic compounds and claimed to have revised some structures and proposed new mechanisms

Total synthesis: an enabling science

• Bastien Nay

Beilstein J. Org. Chem. 2023, 19, 474–476, doi:10.3762/bjoc.19.36

• [12], as illustrated in this thematic issue with the synthesis of pheromones [16]. This requires permanent technological progress. Thus, the recent boom of artificial intelligence, machine learning, and computational chemistry for retrosynthetic analyses and beyond foreshadows a renewed interest in

Navigating and expanding the roadmap of natural product genome mining tools

• Friederike Biermann,
• Sebastian L. Wenski and
• Eric J. N. Helfrich

Beilstein J. Org. Chem. 2022, 18, 1656–1671, doi:10.3762/bjoc.18.178

• developed for the biosynthetic rule-based identification of natural product gene clusters. Apart from these hard-coded algorithms, multiple tools that use machine learning-based approaches have been designed to complement the existing genome mining tool set and focus on natural product gene clusters that

• lack genes with conserved signature sequences. In this perspective, we take a closer look at state-of-the-art genome mining tools that are based on either hard-coded rules or machine learning algorithms, with an emphasis on the confidence of their predictions and potential to identify non-canonical

• algorithms based on hard-coded rules to machine learning (ML)-based approaches with regard to the natural product biosynthetic principles they are most suited for. We focus on how the different genome mining tools identify BGCs and highlight their advantages and limitations. Moreover, we will showcase two

Molecular and macromolecular electrochemistry: synthesis, mechanism, and redox properties

• Shinsuke Inagi and
• Mahito Atobe

Beilstein J. Org. Chem. 2022, 18, 1505–1506, doi:10.3762/bjoc.18.158

• so on, it has a high affinity to informatics approaches, e.g., machine learning, which is expected to become an increasingly important tool in the future. Progress in the design of organic molecules and polymers and the understanding of the redox behavior of these compounds has led to the development

Cytochrome P450 monooxygenase-mediated tailoring of triterpenoids and steroids in plants

• Karan Malhotra and
• Jakob Franke

Beilstein J. Org. Chem. 2022, 18, 1289–1310, doi:10.3762/bjoc.18.135

• combination with ground-breaking machine learning approaches for protein structure prediction such as AlphaFold2 [108], we anticipate that the catalytic repertoire of CYPs will be exploited much more for the biotechnological production of tailor-made triterpenoids and steroids in the near future. We hope that

Molecular basis for protein–protein interactions

• Brandon Charles Seychell and
• Tobias Beck

Beilstein J. Org. Chem. 2021, 17, 1–10, doi:10.3762/bjoc.17.1

• characterisation of the binding reaction. Computational methods are used to predict PPIs and interfaces. The advantage of performing in silico experiments includes narrowing down the number of the binding partners to be tested in vitro or in vivo. Computational methods include supervised machine learning, where

A consensus-based and readable extension of Linear Code for Reaction Rules (LiCoRR)

• Benjamin P. Kellman,
• Yujie Zhang,
• Emma Logomasini,
• Eric Meinhardt,
• Karla P. Godinez-Macias,
• Austin W. T. Chiang,
• James T. Sorrentino,
• Chenguang Liang,
• Bokan Bao,
• Yusen Zhou,
• Sachiko Akase,
• Isami Sogabe,
• Thukaa Kouka,
• Elizabeth A. Winzeler,
• Iain B. H. Wilson,
• Matthew P. Campbell,
• Sriram Neelamegham,
• Frederick J. Krambeck,
• Kiyoko F. Aoki-Kinoshita and
• Nathan E. Lewis

Beilstein J. Org. Chem. 2020, 16, 2645–2662, doi:10.3762/bjoc.16.215

• such models using a variety of strategies, including mechanistic and nonlinear [4][5][6][7][8][9][10][11][12], linear probabilistic [13][14], machine learning [15], formal-grammar [16], and substructural [17]. Unfortunately, most of these approaches use slightly different expressions of the building

Models of necessity

• Timothy Clark and
• Martin G. Hicks

Beilstein J. Org. Chem. 2020, 16, 1649–1661, doi:10.3762/bjoc.16.137

• not always clear to practicing chemists, so that controversial discussions about the merits of alternative models often arise. However, the extensive use of artificial intelligence (AI) and machine learning (ML) in chemistry, with the aim of being able to make reliable predictions, will require that

• molecules suitable for depiction in databases, cheminformatics, machine learning (ML) or artificial intelligence (AI): It is essential for chemists to be able to communicate with each other about molecules. The language of chemistry varies slightly between the organic and inorganic communities. However, it

• reactions are relatively straightforward constructions, if we look further, for example to systems for predicting reactions or suggesting synthetic routes [45], whether using manually coded transformations or developments using automated machine learning and AI techniques, limitations of the Lewis model

In silico rationalisation of selectivity and reactivity in Pd-catalysed C–H activation reactions

• Liwei Cao,
• Mikhail Kabeshov,
• Steven V. Ley and
• Alexei A. Lapkin

Beilstein J. Org. Chem. 2020, 16, 1465–1475, doi:10.3762/bjoc.16.122

• desired. Recent years have seen the emergence of new methods of research in chemistry and process development, which include high-throughput experiments [3], autonomous self-optimising reactors [4][5][6], as well as predictions of reaction outcomes and of reaction conditions based on machine learning (ML

• demonstrated. While machine learning methods are showing great promise and continue to be improved upon, it is also clear that a ML model is unlikely to ever be able to compete in accuracy and interpretability with fully predictive mechanistic models, were it not for the prohibitively high cost of developing

• developing machine learning models for predicting reaction outcomes. C–H activation reactions allow conversion of relatively inexpensive and abundant hydrocarbons into the more sophisticated value-added molecules [11]. With the notion of step-economical and environmentally friendly synthesis, direct

Photophysics and photochemistry of NIR absorbers derived from cyanines: key to new technologies based on chemistry 4.0

• Bernd Strehmel,
• Christian Schmitz,
• Ceren Kütahya,
• Yulian Pang,
• Anke Drewitz and
• Heinz Mustroph

Beilstein J. Org. Chem. 2020, 16, 415–444, doi:10.3762/bjoc.16.40

Bacterial terpene biosynthesis: challenges and opportunities for pathway engineering

• Eric J. N. Helfrich,
• Geng-Min Lin,
• Christopher A. Voigt and
• Jon Clardy

Beilstein J. Org. Chem. 2019, 15, 2889–2906, doi:10.3762/bjoc.15.283

• machine learning and retrobiosynthetic algorithms could facilitate the design of constructs for specific terpenoid variants [149]. While it is now relatively straightforward to direct the flux to produce terpene skeletons, less is known about how to effectively support function of CYPs beyond natural