Machine learning models for rapid prediction of chemicals’ life-cycle environmental impacts: Current status, challenges, and future directions

Abstract

Understanding and reducing the life-cycle environmental impacts of chemicals can benefit a wide range of industries. However, current life cycle assessments (LCA) of chemicals are limited by slow speed and high cost. Molecular-structure-based machine learning (ML) is the most promising technology for the rapid prediction of life-cycle environmental impacts of chemicals. Advances in training datasets, feature engineering, and model architectures for ML models in this area are systematically discussed in this review. We call for the establishment of a large, open, and transparent LCA database for chemicals that includes a wider range of chemical types to address the challenge of data shortages. Greater emphasis on external regulation of data is also needed to produce high-quality LCA data. The construction of more efficient chemical-related descriptors and the identification of features most pertinent to LCA results represent pivotal steps in the advancement of next-generation modeling. The integration of large language models (LLMs) is expected to provide new impetus for database building and feature engineering. Finally, expanding the dimensions of predictable chemical life cycles can further extend the applicability of relevant research. This review aims to outline key strategies to advance this emerging field.