Biosensing with microbial fuel cells and artificial neural networks: laboratory and field investigations

In this study microbial fuel cell-based biosensing was integrated with artificial neural networks (ANNs) in laboratory and field testing of water samples. Inoculation revealed two types of anode-respiring bacteria (ARB) induction profiles, a relatively slow gradual profile and a faster profile that was preceded by a significant lag time. During laboratory testing, the MFCs generated well-organized normally distributed profiles but during field experiments the peaks had irregular shapes and were smaller in magnitude. Generally, the COD concentration correlated better with peak area than with peak height. The ANN predicted the COD concentration (R(2) = 0.99) with one layer of hidden neurons and for concentrations as low as 5 mg acetate-COD/L. Adding 50 mM of 2-bromoethanesulfonate amplified the electrical signals when glucose was the substrate. This report is the first to identify two types of ARB induction profiles and to demonstrate the power of ANNs for interpreting a wide variety of electrical response peaks.