Polymer sequence design via molecular simulation-based active learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Data Science

Background:

• Designing materials at the molecular level is challenging due to vast design spaces.
• Traditional methods struggle to explore these large spaces efficiently.

Purpose of the Study:

• To develop an efficient method for exploring macromolecular sequence space.
• To design target polymer structures using a blended approach.

Main Methods:

• Implemented a blended approach integrating physics-based methods, machine learning, and uncertainty quantification.
• Utilized active learning for sequence optimization of copolymers.
• Assessed the impact of surrogate models, kernels, and initial conditions.

Main Results:

• Demonstrated the efficacy of data-driven methods within active learning for sequence design.
• Identified optimal strategies and hyperparameters for efficient inverse design.
• Successfully screened macromolecular sequence space to design target structures.

Conclusions:

• The active learning framework provides an efficient strategy for inverse design of polymer sequences.
• This approach overcomes limitations of traditional methods in exploring large molecular design spaces.
• Optimal hyperparameters and strategies were established for effective polymer sequence optimization.