A5101455883- Machine Learning in Materials Science
- Spectroscopy and Quantum Chemical Studies
- Computational Drug Discovery Methods
- Spectroscopy Techniques in Biomedical and Chemical Research
- Metabolomics and Mass Spectrometry Studies
- Electronic and Structural Properties of Oxides
- Risk and Safety Analysis
- Nuclear Physics and Applications
- Quantum, superfluid, helium dynamics
- Biosimilars and Bioanalytical Methods
- X-ray Diffraction in Crystallography
- Photoreceptor and optogenetics research
- Spectroscopy and Chemometric Analyses
- SARS-CoV-2 detection and testing
- Fuel Cells and Related Materials
- Advanced NMR Techniques and Applications
- Electron and X-Ray Spectroscopy Techniques
- Laser-Plasma Interactions and Diagnostics
- Machine Learning and ELM
- Protein Structure and Dynamics
- Machine Learning and Data Classification
- Spectroscopy and Laser Applications
- Advanced Chemical Physics Studies
Bayer (Germany)
2020-2023
Stanford University
2019-2023
Bayer (United States)
2022
Ruhr University Bochum
2011
Artificial neural networks represent an accurate and efficient tool to construct high-dimensional potential-energy surfaces based on first-principles data. However, so far the main drawback of this method has been limitation a single atomic species. We present generalization compounds arbitrary chemical composition, which now enables simulations wide range systems containing large numbers atoms. The required incorporation long-range interactions is achieved by combining numerical accuracy...
Over the past years high-dimensional neural network potentials (HDNNPs), fitted to accurately reproduce ab initio potential energy surfaces, have become a powerful tool in chemistry, physics and materials science. Here, we focus on training of networks that lies at heart HDNNP method. We present an efficient approach for optimizing weight parameters via multistream Kalman filtering, using energies forces as reference data. In this procedure, choice free filter can significant impact fit...
Federated multipartner machine learning has been touted as an appealing and efficient method to increase the effective training data volume thereby predictivity of models, particularly when generation is resource-intensive. In landmark MELLODDY project, indeed, each ten pharmaceutical companies realized aggregated improvements on its own classification or regression models through federated learning. To this end, they leveraged a novel implementation extending multitask across partners,...
The excited-state dynamics of chromophores in complex environments determine a range vital biological and energy capture processes. Time-resolved, multidimensional optical spectroscopies provide key tool to investigate these Although theory has the potential decode spectra terms electronic atomistic dynamics, need for large numbers structure calculations severely limits first-principles predictions condensed phase. Here, we leverage locality chromophore excitations develop machine learning...
Abstract Atomistic simulations have become an invaluable tool for industrial applications ranging from the optimization of protein-ligand interactions drug discovery to design new materials energy applications. Here we review recent advances in use machine learning (ML) methods accelerated based on a quantum mechanical (QM) description system. We show how progress ML has dramatically extended applicability range conventional QM-based simulations, allowing calculate industrially relevant...
Federated multi-partner machine learning can be an appealing and efficient method to increase the effective training data volume thereby predictivity of models, particularly when generation is resource intensive. In landmark MELLODDY project, each ten pharmaceutical companies realized aggregated improvements on its own classification and/or regression models through federated learning. To this end, they leveraged a novel implementation extending multi-task across partners, platform audited...
Molecules with an excess number of hydrogen-bonding partners play a crucial role in fundamental chemical processes, ranging from anomalous diffusion supercooled water to transport aqueous proton defects and ordering around hydrophobic solutes. Here we show that overcoordinated hydrogen-bond environments can be identified both the ambient regimes liquid by combining experimental Raman multivariate curve resolution measurements machine learning accelerated quantum simulations. In particular,...
The transport of excess protons and hydroxide ions in water underlies numerous important chemical biological processes. Accurately simulating the associated mechanisms ideally requires utilizing ab initio molecular dynamics simulations to model bond breaking formation involved proton transfer path-integral nuclear quantum effects relevant light hydrogen atoms. These requirements result a prohibitive computational cost, especially at time length scales needed converge properties. Here, we...
Recent advances in machine-learning interatomic potentials have enabled the efficient modeling of complex atomistic systems with an accuracy that is comparable to conventional quantum mechanics based methods. At same time, construction new can seem a daunting task, as it involves data-science techniques are not yet common chemistry and materials science. Here, we provide tutorial-style overview strategies best practices for artificial neural network (ANN) potentials. We illustrate most...
Machine learning potentials (MLPs) trained on data from quantum-mechanics based first-principles methods can approach the accuracy of reference method at a fraction computational cost. To facilitate efficient MLP-based molecular dynamics (MD) and Monte Carlo (MC) simulations, an integration MLPs with sampling software is needed. Here we develop two interfaces that link Atomic Energy Network ({\ae}net) MLP package popular packages TINKER LAMMPS. The three packages, {\ae}net, TINKER, LAMMPS,...
Machine learning models predicting the bioactivity of chemical compounds belong nowadays to standard tools cheminformaticians and computational medicinal chemists. Multi-task federated are promising machine approaches that allow privacy-preserving usage large amounts data from diverse sources, which is crucial for achieving good generalization high-performance results. Using large, real world sets six pharmaceutical companies, here we investigate different strategies averaging weighted task...
Time-resolved scattering experiments enable imaging of materials at the molecular scale with femtosecond time resolution. However, in disordered media they provide access to just one radial dimension thus limiting study orientational structure and dynamics. Here we introduce a rigorous practical theoretical framework for predicting interpreting combining optically induced anisotropy time-resolved scattering. Using impulsive nuclear Raman ultrafast x-ray chloroform simulations, demonstrate...
Molecules with an excess number of hydrogen-bonding partners play a crucial role in fundamental chemical processes, ranging from the anomalous diffusion supercooled water to transport aqueous proton defects and ordering around hydrophobic solutes. Here we show that overcoordinated hydrogen bond environments can be identified both ambient regimes liquid by combining experimental Raman multivariate curve resolution measurements machine learning accelerated quantum simulations. In particular,...
The excited state dynamics of chromophores in complex environments determine a range vital biological and energy capture processes. Time-resolved, multidimensional optical spectroscopies provide key tool to investigate these Although theory has the potential decode spectra terms electronic atomistic dynamics, need for large numbers structure calculations severely limits first principles predictions condensed phase. Here, we leverage locality chromophore excitations develop machine learning...
Machine learning models predicting the bioactivity of chemical compounds belong nowadays to standard tools cheminformaticians and computational medicinal chemists. Multi-task federated are promising machine approaches that allow priva-cy-preserving usage large amount data from diverse sources, which is crucial for achieving good generalization high-performance results. Using large, real world sets six pharmaceutical companies, here we investigate different strategies averaging weighted task...
Machine learning models predicting the bioactivity of chemical compounds belong nowadays to standard tools cheminformaticians and computational medicinal chemists. Multi-task federated are promising machine approaches that allow priva-cy-preserving usage large amount data from diverse sources, which is crucial for achieving good generalization high-performance results. Using large, real world sets six pharmaceutical companies, here we investigate different strategies averaging weighted task...