Towards the design of artificial sensing materials via quantum-informed explainable AI

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Designing sensing materials is challenging due to complex molecular interactions and lack of clear property correlations.
• Reliable molecular recognition is crucial for non-invasive diagnostics in healthcare applications.

Purpose of the Study:

• To develop a quantum-informed AI framework (MORE-ML) for the computational design of artificial sensing materials.
• To uncover sensing mechanisms and guide the design of new electronic-nose (e-nose) systems for body odor volatilome (BOV) analysis.

Main Methods:

• Expanded the MORE-Q dataset to MORE-QX, including conformational analysis of BOV molecules and mucin-derived receptors.
• Utilized quantum-mechanical (QM) properties as inputs for machine learning (ML) models (specifically CatBoost) to predict electronic binding features (BFs).
• Employed explainable AI to identify key descriptors influencing BF predictions.

Main Results:

• Identified weak correlations between QM properties and resulting BFs, prompting a descriptor-based ML approach.
• CatBoost models demonstrated superior performance and transferability to new compounds.
• Reduced high-dimensional QM property space to interpretable descriptors, revealing key influencing factors.

Conclusions:

• MORE-ML integrates QM insights with ML for mechanistic understanding and rational design of BOV sensing materials.
• This framework advances materials for analyzing complex odor mixtures, bridging computational chemistry and practical e-nose applications.