Ab Initio Machine Learning in Chemical Compound Space.

Area of Science:

• Computational chemistry
• Materials science
• Artificial intelligence

Background:

• Chemical compound space (CCS) is vast, making exhaustive virtual sampling computationally prohibitive.
• First-principles calculations, while accurate, are too slow for large-scale exploration of CCS.
• Novel methods are needed to efficiently search CCS for materials with specific properties.

Purpose of the Study:

• To review machine learning (ML) approaches for navigating chemical compound space.
• To highlight Quantum mechanics-based Machine Learning (QML) as a solution for accelerating materials discovery.
• To discuss the integration of quantum mechanics principles into ML models for enhanced accuracy and transferability.

Main Methods:

• Utilizing synthetic data generated from quantum mechanics (QM) based methods.
• Employing ML model architectures inspired by QM principles.
• Combining statistical surrogate models with ab initio physics insights.

Main Results:

• QML approaches offer significant computational speedups compared to traditional QM methods.
• QML models demonstrate universality and transferability across diverse chemical compound space.
• These methods promise to accelerate the discovery of novel molecules and materials.

Conclusions:

• QML represents a powerful paradigm for overcoming the computational bottlenecks in materials science.
• By integrating QM and ML, researchers can efficiently explore vast chemical spaces.
• Future developments in QML are expected to further enhance predictive capabilities and accelerate scientific discovery.