Biocomputing has emerged as a promising field with the potential for energy-efficient computation, but standardized methods to evaluate living systems’ problem-solving capabilities remain underdeveloped. To address this gap, we focus on bacterial systems and their metabolic adaptation. Using an in-silico model of bacterial behavior in varying environments, we propose a new framework for transforming supervised machine learning (ML) problems into a format solvable by biological systems. We then evaluate the framework’s performance against other ML algorithms on standard classification and regression benchmarks. Experimental results show that bacterial metabolic networks often outperform linear methods and rival boosted trees, which are considered state-of-the-art for tabular data. A final ablation study suggests that the system’s computational capacity may stem from its biological components rather than the translation tools used for the learning problem.