Soil forensics predicting acidic pollutants based on 16S rRNA gene of acidophiles and machine learning

Abstract

In industrial settings, events such as explosions, fires, and container breakages can lead to the extensive leakage of acids into the soil environment. As awareness of acidic soil contamination grows, efforts are being made to identify the sources of such pollution to improve soil sustainability. This study aimed to identify HCl, HF, HNO₃, and H₂SO₄ pollution in soil using 16S rRNA gene profiles of acidophiles. Exposure to these pollutants lowered soil pH to 1.8-2.0, causing a decline in proteobacteria and a rise in acidophilic firmicutes, as shown by NGS and T-RFLP analysis. Using this data for clustering analysis, distinct groupings emerged based on the type of acidic contaminant. Although the T-RFLP dataset provided a clearer distinction compared to NGS, pinpointing the specific acidic contaminants with precision remained challenging. The machine learning model using artificial neural networks achieved a 94.4% accuracy in predicting acidic contaminants using the species level NGS data. When utilizing T-RFLP data, it demonstrated an accuracy of 86.9%, showing performance between the genus and family classification levels of NGS. The artificially augmented T-RFLP data further enhanced predictive accuracy. This combined technology of machine learning and molecular microbial detection offers a new approach to soil contamination monitoring.