High-Accuracy Polymer Property Detection via Pareto-Optimized SMILES-Based Deep Learning | JoVE Visualize

Area of Science:

Materials Science
Computational Chemistry
Polymer Informatics

Background:

Conventional polymer design relies on intuition and experience, facing challenges with the vast design space and demand for novel materials.
Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), offers a promising approach for accelerated materials design.
Existing ML and DL methods show potential but require enhanced models for accurate polymer classification and property prediction.

Purpose of the Study:

To design and develop an advanced AI model for polymer property classification using chemical structure inputs.
To enhance polymer informatics by creating a scalable and domain-specific solution for predicting material properties.
To improve upon existing methods by capturing complex chemical dependencies within polymer structures.

Main Methods:

Development of the Simplified Molecular Input Line Entry System Based Polymer Property Detection and Classification Using Pareto Optimization Algorithm (SMILES-PPDCPOA) model.
Integration of a one-dimensional convolutional neural network (1DCNN) with a gated recurrent unit (GRU) for feature extraction and sequence modeling.
Optimization of the 1DCNN-GRU model hyperparameters using the Pareto Optimization Algorithm (POA) for improved performance.

Main Results:

The SMILES-PPDCPOA model achieved an average classification accuracy of 98.66% across eight polymer property classes.
The model demonstrated high precision and recall metrics, indicating robust classification performance.
SMILES-PPDCPOA exhibited superior computational efficiency, completing tasks in 4.97 seconds, outperforming established methods like GCN-LR and ECFP-NN.

Conclusions:

The proposed SMILES-PPDCPOA model offers a novel and effective deep learning framework for polymer property classification.
The integration of 1DCNN, GRU, and Pareto Optimization provides a scalable and accurate solution for polymer informatics.
Experimental validation confirms the potential of SMILES-PPDCPOA as a promising approach for advancing materials science and engineering.