Dataset for simulating biofilm growth

Accurately modeling biofilm growth in porous media is essential for applications in environmental remediation, medical treatment and energy development. However, simulating this process remains challenging due to the intricate interplay between physical and biochemical dynamics. Our study reveals that biofilm growth and diffusion exhibit a complex non-monotonic relationship with biomass, modulated by environmental and intrinsic biological factors. Based on a newly modified density-dependent biofilm diffusion model, a hybrid biofilm growth simulation framework combining microfluidic experiments and a physics-informed neural network (PINN) was developed. Compared to numerical simulations and purely data-driven models, the hybrid model offers more accurate simulations of biofilm growth and diffusion under experimental conditions (with a coefficient of determination of 93%), and successfully inverting the optimal correction parameters for the mechanisms. Utilizing transfer learning, the high generalization capability of PINN was validated in a new dynamic scenario with sparse data. This high-fidelity framework provides a powerful tool for biofilm growth modeling and offers new theoretical insights and strategies for applications in related fields.