Abstract
Understanding the impact of gene deletions is crucial for biological discovery, biomedicine, and biotechnology. Due to the complexity of genome-wide deletion screens, there is growing interest in computational methods that leverage existing screening data for predictive modeling. Here, we present Flux Cone Learning, a general framework designed to predict the effects of metabolic gene deletions on cellular phenotypes. Using Monte Carlo sampling and supervised learning, our approach identifies correlations between the geometry of the metabolic space and experimental fitness scores from deletion screens. Flux Cone Learning delivers best-in-class accuracy for prediction of metabolic gene essentiality in organisms of varied complexity (Escherichia coli, Saccharomyces cerevisiae, Chinese Hamster Ovary cells), outperforming the gold standard predictions of Flux Balance Analysis. We demonstrate the versatility of our approach by training a predictor of small molecule production using data from a large deletion screen. Flux Cone Learning can be applied to many organisms and phenotypes, without the need to encode cellular objectives as an optimization task. Our work offers a broadly applicable tool for phenotypic prediction and lays the groundwork for building metabolic foundation models across the kingdom of life.