🪐 janus-core
High-level Python API and rich CLI for materials modeling with machine-learned interatomic potentials (MACE, SevenNet, CHGNet, M3GNet).
pip install janus-core
STFC • UKRI • PSDI Data to Knowledge • Sci-Tech Daresbury
Data Driven Materials & Molecular Science
Pioneering the convergence of physics-informed machine learning, high-performance molecular dynamics, and automated workflows to decipher materials from quantum to continuum scales.
60+
Publications
4
Core Packages
100%
Open Science
HPC
Scale Science
The Data Driven Materials and Molecular Science (DDMMS) group is based at the Science and Technology Facilities Council (STFC) Scientific Computing Department at Sci-Tech Daresbury, UKRI.
Our research overcomes the traditional trade-off between quantum mechanical accuracy and large-scale simulation capabilities. By creating equivariant graph neural network potentials, automated calculation workflows, and high-performance computing pipelines, we enable predictive simulation of complex materials for energy storage, catalysis, carbon capture, and quantum technologies.
Machine Learning Potentials
Equivariant foundation models (MACE, SevenNet, CHGNet) delivering DFT fidelity across millions of atoms.
Extreme-Scale MD
Massively parallel algorithms, GPU offloading, and symplectic statistical mechanics.
Nanoporous Materials
Metal-organic frameworks (uMOF), negative thermal expansion, phonons, and gas adsorption.
Automated Workflows
Reproducible pipelines with janus-core, aiida-mlip, ml-peg, and goldilocks.
Theme 1
Foundation ML Interatomic Potentials
Developing equivariant graph neural networks (MACE, SevenNet, CHGNet), polarizable electrostatics (MACE-POLAR), and active-learning training set optimization.
Theme 2
Metal-Organic Frameworks & Porous Solids
High-throughput screening of flexible MOF structures, negative thermal expansion (NTE) mechanics, vibrational phonon dynamics, and selective catalytic centers.
Theme 3
Complex Fluids & Molten Salts
Microscopic transport properties, ionic correlations, viscosity, and fundamental bounds of thermal conductivity in molten salts for green energy systems.
🔄 aiida-mlip
AiiDA plugin integrating janus-core for reproducible calculations with machine-learned interatomic potentials and full data provenance.
pip install aiida-mlip
📊 ml-peg
Machine Learning Performance and Extrapolation Guide. A benchmarking framework evaluating MLIPs across diverse systems and physical properties.
git clone https://github.com/ddmms/ml-peg.git
🐻 goldilocks
Web application and library for generating input files with optimised k-point meshes for Quantum ESPRESSO SCF calculations. Part of the PSDI Data to Knowledge initiative.
pip install goldilocks
A budget-dependent crossover between coverage-and response-based training-set selection for machine-learned interatomic potentials
A strategic roadmap for an atomistic machine-learning ecosystem
An Efficient On-the-Fly Nonadiabatic Coupling Framework Integrated into CP2K
Automatic generation of input files with optimised k-point meshes for Quantum ESPRESSO self-consistent field single-point total energy calculations
Dr. Alin Marin Elena
Group Leader • Principal Computational Scientist
STFC SCD, UKRI | CCP5 Scientific Secretary
Specializing in atomistic molecular dynamics, machine-learned interatomic potentials, DL_POLY development, and scientific computing infrastructures.
Elliott Kasoar
Computational Scientist • Research Associate
STFC SCD, UKRI
Specialist in equivariant graph neural networks, foundation interatomic potentials (MACE), active learning, and lead developer of janus-core.
Dr. Junwen Yin
Computational Scientist • Research Associate
STFC SCD, UKRI
Expert in ab initio electronic structure methods, nonadiabatic dynamics, extended CP2K simulations, and materials modeling.