ml-materials-predictor

name	ml-materials-predictor
description	Machine learning skill for nanomaterial property prediction and discovery acceleration
allowed-tools	["Read","Write","Glob","Grep","Bash"]
metadata	{"specialization":"nanotechnology","domain":"science","category":"computational","priority":"high","phase":6,"tools-libraries":["MatMiner","MEGNet","CGCNN","scikit-learn","PyTorch"]}
graph	{"domains":["domain:nanotechnology"],"skillAreas":["skill-area:mathematical-reasoning","skill-area:physics-simulation","skill-area:data-analysis"],"workflows":["workflow:experiment-design"],"roles":["role:research-engineer"]}

ML Materials Predictor

Purpose

The ML Materials Predictor skill provides machine learning capabilities for accelerated nanomaterial discovery and property prediction, enabling data-driven approaches to materials design and optimization.

Capabilities

• Feature engineering for materials
• Property prediction models (GNN, transformers)
• Active learning for experiment design
• High-throughput virtual screening
• Synthesis success prediction
• Transfer learning for small datasets

Usage Guidelines

ML Materials Workflow

1. Data Preparation

   • Collect and curate dataset
   • Generate features (composition, structure)
   • Handle missing values

2. Model Development

   • Select appropriate architecture
   • Train with cross-validation
   • Evaluate on held-out test

3. Application

   • Screen candidate materials
   • Prioritize experiments
   • Validate predictions

Process Integration

• Machine Learning Materials Discovery Pipeline
• Structure-Property Correlation Analysis

Input Schema

{
  "dataset_file": "string",
  "target_property": "string",
  "model_type": "random_forest|gnn|cgcnn|megnet",
  "features": "composition|structure|both",
  "task": "train|predict|screen"
}

Output Schema

{
  "model_performance": {
    "mae": "number",
    "rmse": "number",
    "r2": "number"
  },
  "predictions": [{
    "material": "string",
    "predicted_value": "number",
    "uncertainty": "number"
  }],
  "top_candidates": [{
    "material": "string",
    "predicted_property": "number",
    "rank": "number"
  }]
}