Abstract
Challenges such as antibiotic resistance, ecosystem resilience, and bioproduction optimization require quantitative methods to characterize microbial responses to environmental perturbations. However, translating rapidly growing microbial growth datasets into actionable insights remains challenging. To address this issue, we introduce Kinbiont-an open-source tool that integrates dynamic models with machine learning methods for data-driven discovery in microbiology. Kinbiont consists of three sequential yet independent modules: (1) data preprocessing, (2) model-based parameter inference with both user-defined differential equation systems and hard-coded growth models, and (3) explainable machine learning analyses to map experimental conditions directly to inferred biological parameters. We benchmark Kinbiont using various microbial growth datasets, including diauxic curves, ethanol bioproduction, and phage-bacteria interactions. To illustrate Kinbiont's ability to automatically identify mathematical relationships underlying microbial responses, we revisit Monod's classical nutrient-limitation experiment and perform a growth inhibition assay using a ribosome-targeting antibiotic. In a large-scale ecotoxicological screen, Kinbiont reveals growth-phase-specific sensitivities to environmental stressors. Together, these results demonstrate how Kinbiont converts microbial kinetics data into interpretable and testable hypotheses, acting as a powerful tool to accelerate discovery in modern microbiology.