Deep-Learning-Enhanced Living Biophotovoltaics: Predictive Photocurrent Modeling and Sensitive Herbicide Biosensing.
AI summary
75% confidenceA cyanobacteria-based living biophotovoltaic system was developed for green energy generation and herbicide detection, with optimized parameters for photocurrent output. The system utilized deep learning architectures to model and forecast chronoamperometric photocurrent dynamics. The BiLSTM-SGDM model showed the best predictive performance with high accuracy.
Generated by MESSAI extraction pipeline · review against source PDF
No 3D model is mapped to this paper yet. Parameter ranges above still place reported values on the literature distribution.
What they did
- System
- MSC
What worked
No outcome metrics extracted yet.
Abstract
A cyanobacteria-based living biophotovoltaic (LBPV) system was developed by integrating Leptolyngbya sp. with conductive polymer-gold nanoparticle-modified electrodes for simultaneous green energy generation and herbicide detection. The photoanode was fabricated through the electropolymerization of dithieno[3,2-b:2',3'-d] pyrrole derivatives, followed by the incorporation of aniline-functionalized AuNPs to enhance electron transfer. Optimization of the polymer thickness, AuNP loading, and cyanobacterial concentration revealed 60 electropolymerization cycles and 450 mg/mL cyanobacteria as the ideal parameters for photocurrent output. The biocathode, modified with bilirubin oxidase, enabled efficient oxygen reduction, ensuring stability and reproducibility. To extend the experimental findings, deep learning architectures (LSTM, BiLSTM, and GRU) were employed to model and forecast chronoamperometric photocurrent dynamics. Among all tested configurations, the BiLSTM-SGDM model exhibited the best predictive performance with R 2 = 0.92, RMSE ≈ 48 μA, and MAE ≈ 38 μA on the test set, effectively capturing nonlinear variations and transient response behaviors of the LBPV system. The deep-learning-based predictions closely matched the experimental measurements, confirming the capability of AI-assisted models to reproduce complex photoelectrochemical kinetics. The optimized system produced stable photocurrents under visible light of ∼1 sun (1400 W/m 2 ) and maintained 56% of its initial activity after 50 days. As a biosensor, the LBPV exhibited remarkable sensitivity with detection limits of 1.12 nM for diuron and 9.70 nM for linuron. The integration of AI-based photocurrent forecasting with biohybrid photovoltaic design offers a promising framework for next-generation sustainable energy and environmental monitoring systems. Interference studies further confirmed high selectivity against common environmental contaminants. These findings underscore the potential of LBPVs as dual-function devices, combining sustainable energy harvesting with highly sensitive photoelectrochemical biosensing of phenyl urea herbicides in aquatic environments.