Overview
AlphaCrystal is a contact map-based deep learning algorithm for crystal structure prediction, inspired by AlphaFold's approach to protein structure prediction. AlphaCrystal-II extends this with distance matrix-based prediction.
Theoretical Basis
- Contact map prediction
- Distance matrix prediction (AlphaCrystal-II)
- Deep residual neural networks
- Structure reconstruction from distances
- Composition-to-structure mapping
Key Capabilities
- Crystal structure prediction from composition
- Contact/distance map prediction
- Deep learning approach
- No DFT during prediction
- Fast inference
Sources: ACS Omega (2023), arXiv:2404.04810
Key Strengths
Methodology:
- AlphaFold-inspired
- Contact/distance maps
- Deep learning
Speed:
- Fast inference
- No DFT required
- Large-scale applicable
Innovation:
- Novel approach
- Protein-inspired
- Distance-based
Inputs & Outputs
- Input formats: Chemical composition
- Output data types: Predicted crystal structures, distance maps
Interfaces & Ecosystem
- Framework: TensorFlow/PyTorch
- Tools: MLatticeABC, Cryspnet
- Validation: DFT codes
Workflow and Usage
- Input chemical composition
- Predict distance/contact map
- Reconstruct 3D structure
- Refine structure
- Validate with DFT
Performance Characteristics
- Fast prediction
- GPU-accelerated
- Good for screening
Computational Cost
- Prediction: fast
- No DFT required
- Validation: DFT
Best Practices
- Use for initial screening
- Validate top predictions
- Consider multiple candidates
- Check structural validity
Limitations & Known Constraints
- Training data dependent
- May miss novel structures
- Requires validation
Application Areas
- High-throughput screening
- Initial structure guessing
- Materials discovery
- Composition-to-structure
Comparison with Other Codes
- vs DiffCSP: Different approach (maps vs diffusion)
- vs CSPML: Different methodology
- Unique strength: AlphaFold-inspired, distance maps
Community and Support
- Open-source (GitHub)
- USC development
- Published methodology
Verification & Sources
Primary sources:
- GitHub: https://github.com/usccolumbia/AlphaCrystal
- AlphaCrystal: ACS Omega (2023)
- AlphaCrystal-II: arXiv:2404.04810
Confidence: VERIFIED
Verification status: ✅ VERIFIED
- Website: ACTIVE (GitHub)
- Source: OPEN
- Development: ACTIVE
- Applications: Crystal structure prediction