Interpretable Machine Learning of Two-Photon Absorption.

Area of Science:

• Materials Science
• Computational Chemistry
• Photophysics

Background:

• Strong two-photon absorption (TPA) molecules are crucial for advanced applications like upconverted lasers and photodynamic therapy.
• Current design strategies are limited by the high cost of experimental screening and quantum chemical (QC) calculations.

Purpose of the Study:

• To develop a fast and accurate machine learning (ML) model for predicting TPA magnitudes.
• To identify key molecular features that govern TPA properties through interpretable ML analysis.
• To facilitate high-throughput screening of novel TPA materials.

Main Methods:

• Compiled an experimental TPA database of approximately 900 molecules.
• Utilized interpretable machine learning (ML) to analyze structure-property relationships.
• Developed and validated a predictive ML model against experimental and QC data.

Main Results:

• The ML model achieved prediction errors comparable to experimental and affordable QC methods.
• Identified conjugation length as the most significant feature influencing TPA, followed by donor/acceptor substitution and coplanarity.
• ML analysis quantified and refined existing understanding of TPA-governing molecular features.

Conclusions:

• The developed ML model offers a cost-effective and efficient alternative for TPA material discovery.
• This approach accelerates the design and screening of molecules for TPA-driven applications.
• The findings provide deeper insights into molecular design principles for enhanced two-photon absorption.