Machine learning-based QSAR modeling assay for the nitrification inhibition of 2,4,5-trichloroaniline-derived eco-friendly Schiff bases

Abstract

Twenty-one Schiff bases were synthesized from 2,4,5-trichloroaniline and characterized by instrumental techniques such as ¹H NMR, ¹³C NMR and IR spectroscopy. The compounds were tested for their ability to inhibit nitrification in soil nitrifying bacteria under controlled laboratory conditions over 28 days at three different concentrations (1%, 5%, and 10%), with N-serve and Dicyandiamide (DCD) used as nitrification inhibitors. All the tested compounds showed markedly higher ammonium-N levels (33–185 mg kg⁻¹) and lower nitrate-N levels (23–137 mg kg⁻¹) compared to the control treatment with urea alone. The synthesized compounds were found to be extremely efficient nitrification inhibitors (15–77%) and enhanced soil microbial activities including soil respiration (32.2–34.0 µg CO₂-C per g), dehydrogenase activity (43–46.4 µg TPF per g soil per 24 h) and microbial biomass carbon (537.2–556.3 µg gm⁻¹), compared to other treatments, demonstrating superior microbial activity and soil health. The most important compounds identified from the series were C21, C20, and C3. Pair t-test was performed within the treatment concentration and found that all the treatments differed significantly in terms of mean responses from each other. Statistical and machine leaning models such as stepwise multiple linear regression, artificial neural network (ANN), support vector regression and random forest were used to validate the quantitative structure activity relationship (QSAR) of the synthesized compounds with their nitrification inhibition property. ANN models developed based on feature variables selected from random forest regression were found to be the best fitted models for predicting nitrification inhibition activity at 1%, 5% and 10% of treatment doses.