Isayev Lab

Materials Informatics

Machine learning-accelerated discovery and optimization of advanced functional materials

Materials Informatics

Research Focus

We develop and apply machine learning methods to diverse materials challenges:

  • Polymer design and optimization
  • Crystal structure prediction
  • Property prediction (QSPR)
  • High-throughput computational screening
  • Integration with experimental validation

Polymer Design and Optimization

Reinforcement Learning for Elastomers

We developed reinforcement learning approaches for designing 3D-printable elastomers with superior mechanical properties. Working with polyurethane formulations, our research created a multi-component reward system that guides reinforcement learning agents toward materials exhibiting both high strength and high extensibility—properties that typically trade off against each other.

Achievements

  • Elastomers with more than double the average toughness compared to baseline datasets
  • Twelve formulations achieving both high strength (>10 MPa) and exceptional strain resistance (>200%)
  • Practical formulations suitable for additive manufacturing
  • Experimental validation of computationally designed materials

Applications

  • Medical devices and biocompatible materials
  • Protective equipment and impact resistance
  • Soft robotics and flexible electronics
  • 3D printing and additive manufacturing

Crystal Structure Prediction

Machine learning potentials enable efficient exploration of polymorphic landscapes—identifying stable crystal forms from molecular structure alone:

Methodology

  • AIMNet2 framework for rapid energy evaluation
  • Exploration of thousands of candidate structures
  • Prediction of relative stability and polymorphic hierarchies
  • Mechanical property assessment across crystal forms

Celecoxib Case Study

Work on celecoxib (a pharmaceutical compound) successfully:

  • Reproduced known polymorph stability hierarchies
  • Identified novel hypothetical structures
  • Evaluated mechanical properties (elastic moduli, hardness) across forms
  • Guided experimental polymorph screening

Impact

Crystal structure prediction enables:

  • Selection of optimal solid forms for pharmaceuticals
  • Understanding structure-property relationships
  • Prediction of processing behavior
  • Intellectual property landscaping

Property Prediction (QSPR)

Quantitative structure-property relationships connect molecular/material structure to function:

Target Properties

  • Electronic Properties: Band gaps, conductivity, dielectric constants
  • Thermal Behavior: Glass transition temperatures, thermal conductivity, stability
  • Mechanical Properties: Elastic moduli, hardness, toughness, fracture resistance
  • Environmental Impact: Degradability, toxicity, sustainability metrics

Methodological Approach

Physics-based descriptors derived from quantum chemistry often outperform purely data-driven approaches, enabling:

  • Interpretable models connecting structure to properties
  • Reliable extrapolation to novel chemical space
  • Identification of design principles
  • Integration of domain knowledge

High-Throughput Screening

Computational Workflows

Combining machine learning potentials with automated computational workflows for rapid materials evaluation:

  1. Library Generation: Systematic enumeration of materials space
  2. ML Screening: Rapid property prediction for millions of candidates
  3. Detailed Evaluation: Quantum chemistry for promising candidates
  4. Experimental Validation: Synthesis and testing of top predictions

Experimental Integration

Closed-loop discovery integrating:

  • Computational predictions
  • Automated synthesis (cloud labs, robotic platforms)
  • High-throughput characterization
  • Active learning for iterative improvement

Materials Classes

Our methods apply across diverse materials:

Polymers

  • Elastomers and thermoplastics
  • Conducting polymers
  • Biomedical polymers
  • Polymer composites

Crystalline Materials

  • Pharmaceutical polymorphs
  • Organic semiconductors
  • Metal-organic frameworks
  • Molecular crystals

Functional Materials

  • Battery materials and electrolytes
  • Photovoltaic materials
  • Catalysts and supports
  • Sensors and responsive materials

Design Principles

Materials informatics requires careful consideration of:

Multi-Objective Optimization

Real materials must satisfy multiple competing criteria:

  • Performance properties (mechanical, electronic, optical)
  • Processability and manufacturing constraints
  • Cost and availability of precursors
  • Environmental and safety considerations
  • Stability and durability

Inverse Design

Rather than screening existing materials, inverse design identifies structures meeting target property specifications:

  • Generative models for materials discovery
  • Optimization in chemical space
  • Constraint satisfaction (synthetic accessibility, stability)
  • Multi-property targeting

Computational Tools

Our materials informatics research leverages:

  • AIMNet2: Neural network potentials for rapid property evaluation
  • Active Learning: Efficient exploration of materials space
  • Generative Models: De novo materials design
  • High-Throughput DFT: Quantum chemistry at scale

Vision

We aim to transform materials development:

  • From: Trial-and-error experimental iteration
  • To: Predictive computational design with targeted synthesis
  • Enabling: Rational materials design, accelerated discovery, sustainable innovation

Applications and Impact

Materials informatics enables:

  • Discovery timelines reduced from years to months
  • Exploration of unprecedented chemical space
  • Multi-property optimization previously impossible
  • Integration with autonomous synthesis platforms
  • Sustainable materials with designed end-of-life properties

Future Directions

Emerging priorities in materials informatics:

  • Machine learning for complex multi-component systems
  • Prediction of processing-structure-property relationships
  • Integration of experimental characterization data
  • Autonomous materials discovery platforms
  • Uncertainty-aware predictions for robust design
  • Sustainability and lifecycle assessment integration

Collaborations

This research involves:

  • Experimental materials scientists for validation
  • Industry partners in polymers, pharmaceuticals, and energy
  • National labs and user facilities for characterization
  • Computational resources and HPC centers
  • Additive manufacturing and processing experts

Publications

See our publications page for detailed research findings in materials informatics, including polymer design, crystal structure prediction, and property prediction methodologies.