About REINVENT4
AI-Powered Molecular Generation Platform
Cite This Tool
If you use REINVENT4 in your research, please cite the original paper:
DOI: 10.1186/s13321-024-00812-5
BibTeX
@article{loeffler2024reinvent4,
title={Reinvent 4: Modern AI-driven generative molecule design},
author={Loeffler, Hannes H and He, Jiazhen and Tibo, Alessandro and Janet, Jon Paul and Voronov, Alexey and Mervin, Lewis H and Engkvist, Ola},
journal={Journal of Cheminformatics},
volume={16},
number={1},
pages={20},
year={2024},
publisher={Springer},
doi={10.1186/s13321-024-00812-5}
}
License: Apache 2.0 — Open source, freely available for academic and commercial use. Developed by AstraZeneca's Molecular AI team.
Overview
REINVENT4 is an advanced AI-driven molecular generation platform developed by AstraZeneca's Molecular AI team, integrated with InsilicoΣ for seamless QSAR-guided drug design.
The platform uses deep learning and reinforcement learning algorithms to generate novel molecules optimized for user-defined objectives, enabling rapid exploration of chemical space and identification of promising drug candidates.
System Architecture
Core Components
-
Generative Model Engine
RNN-based molecular generator trained on large chemical databases
-
Reinforcement Learning Module
Optimizes molecules based on multi-objective scoring functions
-
QSAR Integration Layer
Connects with InsilicoΣ QSAR models for activity prediction
-
Scoring Framework
Flexible multi-component scoring with drug-likeness filters
-
Molecule Analysis Pipeline
Automatic property calculation and visualization
Key Principles
Design Philosophy
-
AI-Driven Design
Deep learning models learn chemical patterns and generate novel structures
-
Multi-Objective Optimization
Simultaneously optimize multiple properties (activity, ADMET, synthesizability)
-
Chemical Space Exploration
Efficiently navigate vast chemical space to find optimal molecules
-
QSAR-Guided Generation
Use your own QSAR models to guide molecule generation
-
Drug-likeness by Design
Built-in filters ensure generated molecules are synthetically accessible
Generation Workflow
Define Objectives
Select target QSAR model and configure scoring functions. Define optimization criteria (similarity, drug-likeness, specific properties).
Configure Generation
Set parameters: number of molecules to generate, diversity settings, similarity thresholds, and filter criteria.
AI Generation Process
REINVENT4 generates molecules using reinforcement learning, iteratively improving based on scoring function feedback.
Property Prediction
Generated molecules are automatically scored using selected QSAR models and property calculators.
Analysis & Ranking
Molecules are analyzed, ranked by composite scores, and clustered by chemical similarity for easy review.
Export & Synthesis
Export top candidates in multiple formats (SDF, CSV, SMILES) for further validation or synthesis planning.
Key Features
Deep Learning Generation
State-of-the-art RNN models trained on millions of molecules
QSAR Integration
Direct integration with your InsilicoΣ QSAR models
Smart Filtering
Drug-likeness, PAINS, and custom property filters
Real-time Monitoring
Track generation progress with live statistics and plots
Molecule Bookmarking
Mark favorites and add notes for promising candidates
Multi-format Export
Export as SDF, CSV, Excel with full property data
Common Use Cases
Lead Optimization
Generate optimized analogs of lead compounds with improved activity and ADMET properties
De Novo Design
Create entirely novel chemical scaffolds targeting specific biological activity
Chemical Space Exploration
Systematically explore chemical space around active molecules to find new chemotypes
Multi-Parameter Optimization
Simultaneously optimize multiple properties (potency, selectivity, solubility, toxicity)