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

• as neural networks and support vector machines, to fit models to chemical reaction data that were then optimized by genetic algorithms [78][79][80]. However, the use of ML for chemical reaction optimization did not become popular until the introduction of BO techniques by Lapkin and Bourne et al. [81

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

• benchmarked in various reaction-related tasks [124][125][126][127][128], and has become one of the mainstream reaction-level featurization techniques. Graph-based representation Graph neural networks (GNNs) have been widely applied to various chemical tasks, such as predicting molecular properties [129][130

• SMILES and are instead described using textual labels. A comparison of three types of reaction embedding methods: (A) descriptor-based, which use predefined features from reactants and products, (B) graph-based, which use neural networks to learn features from molecular graphs, and (C) text-based, which

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

• is challenging. Some of the most widely used algorithms include multivariate linear regression (MLR) [69] in which the target is linearly modelled by multiple independent variables. Other notable architectures include decision trees [70], support vector machines [67] and deep neural networks [71][72

• addition, such larger data sets also lead to an increased interest in the application of deep learning tools, such as graph-based neural networks, to organocatalysis. One particular example was published by Hong and co-workers [113], who developed a chemistry-informed graph model for the prediction of

• inspired a manifold of other groups to develop new ML techniques, including graph neural networks. With the continued rise of high-throughput experimentation in organocatalysis [40], we expect ML to be applied to more data sets in this domain to aid in answering a wider variety of research questions. For

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

• training data expansion that allows the more effective usage of deep neural networks for the identification of the most potent compounds from limited training data until almost all hits in the learning set are selected. A slightly different pattern emerged when comparing the tree-based implementations. AD

• S1) and at 200 iterations (88% vs. 86%, +0.8 leads per dataset on average, p = 0.02, Supporting Information File 1, Table S1). While this difference was not nearly as stark as for the deep neural networks, the identification of an additional lead per project might still provide tangible benefits in

• performance for paired approaches. For example, it has been shown that similarity-based pairing during training compound generation for Siamese neural networks can significantly improve model efficiency [36]. Additionally, active learning-based subsampling is an autonomous and adaptive approach that has been

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

• to RFUs from untested glycans (https://github.com/shauseth/glynet). 3. LectinOracle [108]: it is a freely available deep learning-based model that combines transformer-based representations for proteins and graph convolutional neural networks for glycans to predict their interaction (https

• range of carbohydrates, their derivatives and cyclodextrins, and a specific model PS-S for important carbohydrate monomers. All of these features are available for free without registration as online tools (https://pesto.epfl.ch/). 2. GlyNet [107]: it is a free deep learning algorithm, based on neural

• networks (NN), that allows the user to predict protein-glycan binding. Taking a glycan structure as input, this model is able to predict the strength of the interaction based on the relative fluorescence units (RFUs) measured in the Consortium for Functional Glycomics glycan arrays and extrapolating these

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

• probabilistic deep neural network (Bayesian deep neural networks [12]) to support automated monosaccharide recognition for carbohydrate sequencing. We obtained a highly performing algorithm that we called "GlAIcomics", specifically trained on carbohydrates. Methodology Data production Our carbohydrate analysis

Biomimetic molecular design tools that learn, evolve, and adapt

• David A Winkler

Beilstein J. Org. Chem. 2017, 13, 1288–1302, doi:10.3762/bjoc.13.125

• methods and their potential impacts in chemistry, engineering, and medicine. Keywords: automated chemical synthesis; deep learning; evolutionary algorithms; in silico evolution; machine learning; materials design and development; neural networks; Introduction There is still not a clear understanding of

• algorithm is then discussed and its performance compared to traditional ‘shallow’ neural networks is described in the context of mathematical theorem governing the performance of neural networks. The paper then discusses another very important concept in life and in silico learning, feature selection

• , hardness, credit worthiness etc.). They include artificial neural networks, decision trees and several other types of biologically inspired computational algorithms. They have been applied to most areas of science and technology and have made important contributions to chemistry and related molecular and

Computational methods in drug discovery

• Sumudu P. Leelananda and
• Steffen Lindert

Beilstein J. Org. Chem. 2016, 12, 2694–2718, doi:10.3762/bjoc.12.267

• structure corresponding to the target sequence is known as fold recognition and has been used in structure-based drug discovery studies [48]. GenTHREADER is a popular fold recognition program that uses neural networks for the evaluation of the alignments [52]. MUSTER is a freely available webserver that