POPs identification using simple low-code machine learning.

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Identifying persistent organic pollutants (POPs) from large chemical datasets is challenging.
• Existing machine learning models often require advanced programming and computing resources.
• Accessible tools are needed for effective POP screening.

Purpose of the Study:

• To develop and evaluate machine learning models for POP screening using the PyCaret package.
• To compare PyCaret model performance against a deep convolutional neural network (DCNN).
• To assess the utility of low-code machine learning for environmental science applications.

Main Methods:

• Utilized PyCaret, a Python package, for constructing machine learning models.
• Employed 2D molecular descriptors for POP identification.
• Compared several PyCaret models, including Light Gradient Boosting Machine (LGBM), against a DCNN.

Main Results:

• PyCaret models achieved superior or comparable performance to the DCNN.
• The LGBM model demonstrated high accuracy (96.20%), AUC (97.70%), and F1 score (82.58%).
• The LGBM model effectively identified POPs in REACH PBT and industrial chemical lists.

Conclusions:

• PyCaret offers a simple, accessible approach to building effective POP screening models.
• Low-code machine learning tools like PyCaret empower non-computing professionals in environmental science.
• These tools facilitate prompt assessment and management of chemical substances.