AFLOW-ML

Official Resources

• Homepage: http://aflow.org/aflow-ml/
• Documentation: http://aflow.org/aflow-ml/
• Source Repository: Part of AFLOW codebase
• License: GPL v3

Overview

AFLOW-ML is a machine learning API and library integrated into the AFLOW framework. It provides access to pre-trained machine learning models for predicting materials properties (electronic, thermal, mechanical) based on crystal structure and composition. It allows users to screen materials without performing expensive DFT calculations.

Scientific domain: Materials informatics, machine learning, property prediction
Target user community: Materials scientists, high-throughput researchers

Capabilities (CRITICAL)

• Property Prediction: Predicts band gap, bulk/shear modulus, Debye temperature, heat capacity, thermal conductivity, etc.
• Models: Uses Gradient Boosting Decision Trees (GBDT), Voronoi tessellation features, and PLMF (Property Labeled Materials Fragments).
• API: REST API for programmatic access to predictions.
• Online Interface: Web-based predictor.

Sources: AFLOW-ML website, Sci. Rep. 7, 10766 (2017)

Inputs & Outputs

• Input formats: POSCAR, Composition, Structure JSON
• Output data types: Predicted property values (JSON)

Interfaces & Ecosystem

• AFLOW: Integrated with the main AFLOW framework.
• Python: API client available.

Workflow and Usage

1. Prepare structure (POSCAR).
2. Send to API:

   curl -X POST -d @POSCAR http://aflow.org/API/aflow-ml/v1.0/plmf/v1.0/
   

3. Receive JSON response with predictions.

Performance Characteristics

• Extremely fast (milliseconds per prediction)
• Accuracy depends on training set coverage (AFLOW database)

Application Areas

• Rapid screening of millions of compounds
• Guiding DFT calculations
• Discovery of superhard materials

Community and Support

• Developed by Curtarolo Group (Duke)
• Active development

Verification & Sources

Primary sources:

1. Homepage: http://aflow.org/aflow-ml/
2. Publication: O. Isayev et al., Nat. Commun. 8, 15679 (2017)

Confidence: VERIFIED

Verification status: ✅ VERIFIED

• Website: ACTIVE
• Documentation: AVAILABLE
• Source: OPEN (GPL)
• Development: ACTIVE
• Applications: Machine learning, property prediction