A5018079613- Machine Learning in Materials Science
- Computational Drug Discovery Methods
- Zeolite Catalysis and Synthesis
- Protein Structure and Dynamics
- Fuel Cells and Related Materials
- Electrocatalysts for Energy Conversion
- Electrochemical Analysis and Applications
- Chemical Reaction Mechanisms
- Various Chemistry Research Topics
- Catalysis and Oxidation Reactions
- Chemical Synthesis and Analysis
- Advanced Battery Materials and Technologies
- Metal-Organic Frameworks: Synthesis and Applications
- Molecular Junctions and Nanostructures
- Innovative Microfluidic and Catalytic Techniques Innovation
- RNA and protein synthesis mechanisms
- Organic Light-Emitting Diodes Research
- Advanced battery technologies research
- Topic Modeling
- Polyoxometalates: Synthesis and Applications
- DNA and Nucleic Acid Chemistry
- Advanced biosensing and bioanalysis techniques
- Catalytic Processes in Materials Science
- Photochromic and Fluorescence Chemistry
- Spectroscopy and Quantum Chemical Studies
Massachusetts Institute of Technology
2018-2025
IIT@MIT
2024
Moscow Institute of Thermal Technology
2021-2023
Microsoft Research (United Kingdom)
2022
Robert Bosch (Germany)
2022
Harvard University
2014-2020
University of Zurich
2020
Harvard University Press
2015-2019
Universidad de Salamanca
2006-2014
Heriot-Watt University
2012-2013
We introduce a convolutional neural network that operates directly on graphs. These networks allow end-to-end learning of prediction pipelines whose inputs are graphs arbitrary size and shape. The architecture we present generalizes standard molecular feature extraction methods based circular fingerprints. show these data-driven features more interpretable, have better predictive performance variety tasks.
We report a method to convert discrete representations of molecules and from multidimensional continuous representation. This model allows us generate new for efficient exploration optimization through open-ended spaces chemical compounds. A deep neural network was trained on hundreds thousands existing structures construct three coupled functions: an encoder, decoder predictor. The encoder converts the representation molecule into real-valued vector, these vectors back molecular...
A philosophy for defining what constitutes a virtual high-throughput screen is discussed, and the choices that influence decisions at each stage of computational funnel are investigated, including an in-depth discussion generation molecular libraries. Additionally, we provide advice on storing, analysis, visualization data basis extensive experience in our research group.
Anthraquinone derivatives are being considered for large scale energy storage applications because of their chemical tunability and rapid redox kinetics. The authors investigate four anthraquinone as negative electrolyte candidates an aqueous quinone‐bromide flow battery: anthraquinone‐2‐sulfonic acid (AQS), 1,8‐dihydroxyanthraquinone‐2,7‐disulfonic (DHAQDS), alizarin red S (ARS), 1,4‐dihydroxyanthraquinone‐2,3‐dimethylsulfonic (DHAQDMS). standard reduction potentials all lower than that...
Abstract Molecular dynamics simulations provide theoretical insight into the microscopic behavior of condensed-phase materials and, as a predictive tool, enable computational design new compounds. However, because large spatial and temporal scales thermodynamic kinetic phenomena in materials, atomistic are often computationally infeasible. Coarse-graining methods allow larger systems to be simulated by reducing their dimensionality, propagating longer timesteps, averaging out fast motions....
The stability limits of quinones, molecules that show promise as redox-active electrolytes in aqueous flow batteries, are explored for a range backbone and substituent combinations with high-throughput virtual screening.
Abstract Machine learning (ML) outperforms traditional approaches in many molecular design tasks. ML models usually predict properties from a 2D chemical graph or single 3D structure, but neither of these representations accounts for the ensemble conformers that are accessible to molecule. Property prediction could be improved by using conformer ensembles as input, there is no large-scale dataset contains graphs annotated with accurate and experimental data. Here we use advanced sampling...
Selectivity control in zeolite synthesis Zeolites are widely used many industrial applications, but despite decades of research, their still relies on trial-and-error approaches. Complex nucleation mechanisms and topological diversity lead to strong phase competition, complicating the issue rational design synthesis. Using atomistic simulations, literature mining, human-computer interaction, synthesis, characterization, Schwalbe-Koda et al . developed a computational strategy that enables...
Molecular dynamics (MD) simulation techniques are widely used for various natural science applications. Increasingly, machine learning (ML) force field (FF) models begin to replace ab-initio simulations by predicting forces directly from atomic structures. Despite significant progress in this area, such primarily benchmarked their force/energy prediction errors, even though the practical use case would be produce realistic MD trajectories. We aim fill gap introducing a novel benchmark suite...
Solid polymer electrolytes (SPEs) have the potential to improve lithium-ion batteries by enhancing safety and enabling higher energy densities. However, SPEs suffer from significantly lower ionic conductivity than liquid solid ceramic electrolytes, limiting their adoption in functional batteries. To facilitate more rapid discovery of high SPEs, we developed a chemistry-informed machine learning model that accurately predicts SPEs. The was trained on SPE data hundreds experimental...
High-throughput data generation methods and machine learning (ML) algorithms have given rise to a new era of computational materials science by the relations between composition, structure, properties exploiting such for design. However, build these connections, must be translated into numerical form, called representation, that can processed an ML model. Data sets in vary format (ranging from images spectra), size, fidelity. Predictive models scope interest. Here, we review...
A closed-loop, autonomous molecular discovery platform driven by integrated machine learning tools was developed to accelerate the design of molecules with desired properties. We demonstrated two case studies on dye-like molecules, targeting absorption wavelength, lipophilicity, and photooxidative stability. In first study, experimentally realized 294 unreported across three automatic iterations design-make-test-analyze cycles while exploring structure-function space four rarely reported...
Abstract Massive scale, in terms of both data availability and computation, enables important breakthroughs key application areas deep learning such as natural language processing computer vision. There is emerging evidence that scale may be a ingredient scientific learning, but the importance physical priors domains makes strategies benefits scaling uncertain. Here we investigate neural-scaling behaviour large chemical models by varying model dataset sizes over many orders magnitude,...
It has been challenging to find stable blue organic light emitting diodes (OLEDs) that rely on thermally activated delayed fluorescence (TADF). Lack of host materials well‐fitted the TADF emitters is one critical reasons. The most popular for TADF, bis[2‐(diphenylphosphino)phenyl] ether oxide (DPEPO), leads unrealistically high maximum external quantum efficiency. DPEPO however an unstable material and a poor charge transporting ability, which in turn induces intrinsic short OLED operating...
Organic structure directing agents (OSDAs) play a crucial role in the synthesis of micro- and mesoporous materials especially case zeolites. Despite wide use OSDAs, their interaction with zeolite frameworks is poorly understood, researchers relying on heuristics or computationally expensive techniques to predict whether an organic molecule can act as OSDA for certain zeolite. In this paper, we undertake data-driven approach unearth generalized OSDA–zeolite relationships using comprehensive...
Neural network (NN) interatomic potentials provide fast prediction of potential energy surfaces, closely matching the accuracy electronic structure methods used to produce training data. However, NN predictions are only reliable within well-learned domains, and show volatile behavior when extrapolating. Uncertainty quantification can flag atomic configurations for which confidence is low, but arriving at such uncertain regions requires expensive sampling phase space, often using atomistic...
ConspectusDesigning new materials is vital for addressing pressing societal challenges in health, energy, and sustainability. The combination of physicochemical laws empirical trial error has long guided material design, but this approach limited by the cost experiments difficulty deriving complex guiding principles. space hypothetical to be considered incredibly large, only a small fraction possible compounds can ever tested experimentally. computational techniques atomistic simulation...