A Method for Inferring Polymers Based on Linear Regression and Integer Programming.

Area of Science:

• Computational Chemistry
• Materials Science
• Polymer Science

Background:

• A novel framework combining artificial neural networks (ANNs) and mixed integer linear programming (MILP) has been proposed for chemical compound design.
• Existing methods may not fully address the complexities of polymer structure inference and property prediction.

Purpose of the Study:

• To develop a new method for inferring polymer structures based on the established ANN-MILP framework.
• To introduce novel representations and descriptors for polymers within this framework.
• To evaluate the performance of linear regression within the framework for predicting polymer properties.

Main Methods:

• Representing polymers as monomers and defining new polymer-specific structural descriptors.
• Integrating linear regression as a predictive modeling component within the ANN-MILP framework.
• Conducting computational experiments to assess prediction accuracy and inference capabilities.

Main Results:

• Linear regression models demonstrate effective prediction for certain polymer chemical properties.
• The proposed method successfully infers polymer structures with monomers containing up to 50 non-hydrogen atoms.
• Validation of the novel polymer representation and descriptor approach.

Conclusions:

• The developed method effectively extends the novel framework for polymer design and inference.
• Linear regression is a viable component for predicting specific polymer properties within this computational approach.
• The method shows promise for designing complex polymers with tailored characteristics.