Density Prediction Models for Energetic Compounds Merely Using Molecular Topology.

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Accurate density prediction is crucial for energetic materials, but current methods are time-consuming.
• High-throughput screening requires faster and more efficient property prediction models.

Purpose of the Study:

• To develop accurate and efficient machine learning models for predicting the density of energetic compounds.
• To establish direct structure-density mappings using molecular topology.

Main Methods:

• Three machine learning models were proposed: Support Vector Machine (SVM), Random Forest (RF), and Graph Neural Network (GNN).
• Models were trained and tested on a dataset of over 2000 nitro compounds.
• Density Functional Theory Quantitative Structure-Property Relationship (DFT-QSPR) was used as a benchmark.

Main Results:

• The GNN model achieved 88% accuracy (prediction error < 5%), outperforming SVM (48%), RF (63%), and DFT-QSPR (85%).
• GNN effectively mapped molecular structure to density, surpassing traditional methods in accuracy and efficiency.
• SVM and RF models using only fingerprint bit vectors showed limited success.

Conclusions:

• Graph Neural Networks combined with molecular topology offer a highly accurate and computationally efficient approach for density prediction.
• The GNN-based model is well-suited for high-throughput screening of energetic materials, improving materials design processes.