Application of radial basis function neural network to predict soil sorption partition coefficient using topological...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Machine Learning

Background:

• The soil sorption partition coefficient (logKoc) is crucial for assessing organic chemical environmental risks.
• Accurate prediction of logKoc is needed for diverse and novel compounds.

Purpose of the Study:

• To develop a fast and accurate model for predicting the soil sorption partition coefficient (logKoc).
• To utilize a radial basis function neural network (RBFNN) for logKoc prediction.

Main Methods:

• A RBFNN model was developed using eight topological descriptors from 800 diverse organic compounds.
• Generalized Regression Neural Network (GRNN) was employed within the RBFNN for rapid adaptation.
• The dataset was divided into 560 compounds for training and 240 for testing.

Main Results:

• The RBFNN model demonstrated excellent performance with high correlation coefficients (R2) of 0.995 for training and 0.933 for testing.
• Low root-mean square errors (RMSE) of 0.2321 (training) and 0.413 (testing) were achieved.
• The model showed high accuracy for both known and potentially new chemical compounds.

Conclusions:

• The developed RBFNN model provides a reliable and efficient method for predicting logKoc.
• This predictive capability is valuable for environmental risk assessment of existing and emerging organic chemicals.
• The model's adaptability makes it suitable for predicting logKoc for new market products annually.