High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning

Grosnit, Antoine; Tutunov, Rasul; Maraval, Alexandre Max; Griffiths, Ryan‐Rhys; Cowen-Rivers, Alexander I.; Yang, Lin; Zhu, Lin; Lyu, Wenlong; Chen, Zhitang; Wang, Jun; Peters, Jan; Bou-Ammar, Haitham

doi:10.48550/arxiv.2106.03609

Cited by 20 publications

(22 citation statements)

References 59 publications

(75 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, as an expression for the van der Waals interaction, it would be possible to use the Hamaker formula parameterized with a learnable Hamaker constant. 80,81 Additionally, our model could serve as a reliable oracle for the molecular generative models, 82,83 in case the design objective is focused on the target binding affinity.…”

Section: Discussionmentioning

confidence: 99%

PIGNet: a physics-informed deep learning model toward generalized drug–target interaction predictions

et al. 2022

View full text Add to dashboard Cite

show abstract

Section: Discussionmentioning

confidence: 99%

PIGNet: a physics-informed deep learning model toward generalized drug–target interaction predictions

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Currently, the use of Bayesian optimization techniques for solving problems of global geometry optimization and conformation search is popular, since they allow to find minima for small to medium-sized molecules quickly. [26][27][28][29][30] The similar approach, tree-structured Parzen estimator 31,32 (parzen), was tested in this work. We used the implementation of this method available in the Hyperopt package.…”

Section: Methodsmentioning

confidence: 99%

Tree Parzen estimator for global geometry optimization: A benchmark and database of experimental gas‐phase structures of organic molecules

Andreadi

Zankov²,

Karpov

et al. 2022

J Comput Chem

View full text Add to dashboard Cite

Finding global and local minima on the potential energy surface is a key task for most studies in computational chemistry. Having a set of possible conformations for chemical structures and their corresponding energies, one can judge their chemical activity, understand the mechanisms of reactions, describe the formation of metal‐ligand and ligand‐protein complexes, and so forth. Despite the fact that the interest in various minima search algorithms in computational chemistry arose a while ago (during the formation of this science), new methods are still emerging. These methods allow to perform conformational analysis and geometry optimization faster, more accurately, or for more specific tasks. This article presents the application of a novel global geometry optimization approach based on the Tree Parzen Estimator method. For benchmarking, a database of small organic molecule geometries in the global minimum conformation was created, as well as a software package to perform the tests.

show abstract

“…Recent work has shown that it is possible to set up a VAE with as little as 2500 molecules (training for 30 h on a single GPU) which can achieve comparable accuracy on predicting log P ð Þ to VAEs which use hundreds of thousands of data points. 78 Of course, using a smaller training dataset also shrinks the domain of applicability. We would expect chemical VAEs to continue developing, requiring smaller datasets and less training time without sacrificing domain of applicability, and hope that the questions above will reduce the barriers to entry for new researchers interested in chemical VAEs.…”

Section: Areas Of Further Investigation For Chemical Vaesmentioning

confidence: 99%

A review of molecular representation in the age of machine learning

Wigh

Goodman

Lapkin

2022

WIREs Comput Mol Sci

134

View full text Add to dashboard Cite

Research in chemistry increasingly requires interdisciplinary work prompted by, among other things, advances in computing, machine learning, and artificial intelligence. Everyone working with molecules, whether chemist or not, needs an understanding of the representation of molecules in a machinereadable format, as this is central to computational chemistry. Four classes of representations are introduced: string, connection table, feature-based, and computer-learned representations. Three of the most significant representations are simplified molecular-input line-entry system (SMILES), International Chemical Identifier (InChI), and the MDL molfile, of which SMILES was the first to successfully be used in conjunction with a variational autoencoder (VAE) to yield a continuous representation of molecules. This is noteworthy because a continuous representation allows for efficient navigation of the immensely large chemical space of possible molecules. Since 2018, when the first model of this type was published, considerable effort has been put into developing novel and improved methodologies. Most, if not all, researchers in the community make their work easily accessible on GitHub, though discussion of computation time and domain of applicability is often overlooked.Herein, we present questions for consideration in future work which we believe will make chemical VAEs even more accessible.

show abstract

High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning

Cited by 20 publications

References 59 publications

PIGNet: a physics-informed deep learning model toward generalized drug–target interaction predictions

PIGNet: a physics-informed deep learning model toward generalized drug–target interaction predictions

Tree Parzen estimator for global geometry optimization: A benchmark and database of experimental gas‐phase structures of organic molecules

A review of molecular representation in the age of machine learning

Contact Info

Product

Resources

About