Molecular Foundation Models for Predicting Self-Assembly in Aqueous Mixtures.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Foundation models offer novel molecular representations but are limited to single-component predictions.
• Predicting properties of multi-component mixtures remains a significant challenge in chemistry and materials science.

Purpose of the Study:

• To develop and apply a novel method using foundation models for predicting properties of complex amphiphile mixtures.
• To evaluate the performance of these models against existing methods and validate their predictive power through experimentation.

Main Methods:

• A concentration-weighted average of latent representations from graph neural network and transformer-based molecular foundation models was developed.
• Random forest and feed-forward neural network models were trained using these representations to predict critical micelle concentrations, biphasic separation, and liposome formation.
• High-throughput automated experiments were employed for validation and guiding liposome design.

Main Results:

• The developed models achieved predictive performance comparable to or exceeding prior state-of-the-art methods, including bespoke graph neural networks and physicochemical features.
• The models demonstrated the ability to accurately predict properties for single surfactants, binary mixtures, and complex 7-component amphiphile systems.
• Experimental validation confirmed the models' capacity to extrapolate to novel surfactant mixtures and guide liposome formulation.

Conclusions:

• Foundation models, when adapted with concentration-weighted averaging, provide powerful representations for predicting multi-component mixture properties.
• This approach offers a significant advancement over traditional methods, enabling more efficient exploration and design of amphiphile-based materials.
• The validated models can accelerate the discovery and optimization of new surfactants and liposome formulations.