Gene‐Informed Modeling of Denitrification Process in Groundwater Through Dynamic Flux Balance Analysis and Deep Learning
TL;DR
Incorporating microbial genomic data into groundwater denitrification models via dynamic flux balance analysis and deep learning accurately predicts nitrate depletion and nitrite transients, outperforming conventional geochemistry-only models.
Problem / question
Conventional reactive transport models for groundwater denitrification rely solely on geochemical data, ignoring the explicit microbial genetic and metabolic pathways that drive these biogeochemical reactions.
Methods
The authors developed two models: a mechanistic framework coupling dynamic flux balance analysis (DFBA) with the PFLOTRAN reactive transport model (RTM), and a gene-informed deep-learning model using environmental covariates and functional gene abundances. These were tested using controlled batch and column experiments with alternating surface-water and groundwater flow regimes.
Key findings
Both the DFBA-RTM and the gene-informed deep learning model successfully reproduced observed nitrate (NO3-) depletion and nitrite (NO2-) transients in batch and column experiments. Both gene-informed approaches outperformed conventional RTMs that were constrained only by geochemistry. The deep learning model achieved these predictions at an orders-of-magnitude reduced computational cost compared to the mechanistic model.
Why it matters
This dual approach proves that integrating microbial genomic information improves hydro-biogeochemical modeling, providing both mechanistic insights into pathway-level controls and computationally efficient surrogates for large-scale multi-scenario forecasting.
Limitations
The abstract does not explicitly detail specific limitations, biases, or constraints of the models beyond noting that the mechanistic DFBA-RTM approach has a higher computational cost than the deep learning surrogate.
Takeaway
Combining microbial genomic data with reactive transport modeling and deep learning creates highly accurate, scalable simulations of groundwater denitrification that outperform traditional geochemistry-based models.