What is Forerunner 5 (fr5)?¶

Forerunner 5 is a comprehensive Python toolkit for chemical informatics and optimization, developed as part of the AIchemy project at Imperial College London. The package addresses the critical need for accelerating the translation of predicted materials and reactions from computational models to real-world applications.

Project Overview¶

The Forerunner 5 project focuses on "Identifying and Optimising Reaction Mechanisms" using advanced computational methods. Our goal is to develop an automated platform for reaction mechanism discovery that combines:

Multi-fidelity ML models that integrate computational and experimental data
Advanced optimization algorithms for efficient experimental design
Automated property prediction for chemical systems
Scalable workflows for high-throughput discovery

fr5 Overview

Core Capabilities¶

🧪 Molecular Descriptors¶

Generate comprehensive descriptors for various chemical systems:

Ionic Liquids (ILs): Fragment-based and property-based descriptors
Metal-Organic Frameworks (MOFs): Structural and topological descriptors
Molecules: RDKit-based molecular fingerprints and properties

📊 Property Prediction¶

State-of-the-art machine learning models for predicting chemical properties:

Ionic Liquid Properties¶

Viscosity prediction using CatBoost and neural networks
Density estimation with temperature-dependent models
CO₂ solubility prediction for green chemistry applications

MOF Properties¶

CO₂ uptake capacity for carbon capture applications
Pore volume and surface area characterization
Adsorption isotherms prediction

Reaction Properties¶

Reaction conversion prediction
Selectivity optimization for multi-product reactions
Mechanistic pathway analysis

🎯 Bayesian Optimization¶

Advanced optimization framework for experimental design:

Multi-objective optimization with Pareto front analysis
High-dimensional Gaussian Processes with specialized kernels
Adaptive acquisition functions for efficient exploration
VAE-enabled optimization for categorical variables
Batch optimization with constant variables

Key Features¶

✨ User-Friendly APIs¶

# Simple property prediction
from fr5.property_prediction_il import predict_viscosity

result = predict_viscosity({
    'cation': 'EMIM',
    'anion': 'BF4', 
    'temperature': 298.15
})

⚡ High Performance¶

Optimized algorithms for large-scale optimization
GPU acceleration for neural network models
Efficient caching and memory management
Scalable to thousands of experiments

🔬 Scientific Rigor¶

Validated on experimental datasets
Cross-validated model performance metrics
Uncertainty quantification for predictions
Reproducible workflows and random seeds

🔧 Extensible Design¶

Modular architecture for easy extension
Plugin system for custom models
Flexible data I/O with multiple formats
Database integration for experiment tracking

Integration with AIchemy Ecosystem¶

fr5 is designed to integrate seamlessly with other AIchemy tools and workflows:

Data management: SQLite and Excel integration for experiment tracking
Visualization: Built-in plotting with matplotlib and plotly
Parallel computing: Support for distributed optimization
Cloud deployment: Docker containers and cloud-ready APIs

Target Applications¶

The fr5 toolkit is particularly valuable for:

🏭 Industrial Applications¶

Process optimization in chemical manufacturing
Catalyst discovery and optimization
Solvent selection for green chemistry
Reaction condition optimization

🔬 Academic Research¶

High-throughput screening of chemical libraries
Mechanistic studies of complex reactions
Material property prediction and design
Multi-objective optimization studies

💡 Innovation Acceleration¶

Rapid prototyping of new chemical processes
Data-driven discovery of novel materials
Automated experimental design
Knowledge extraction from chemical databases