Curation of Gene-Protein-Reaction Associations Enhances the iPichia Genome-Scale Metabolic Model of Komagataella phaffii.

Komagataella phaffii is a commercially valuable yeast widely used in biotechnology due to its Crabtree-negative metabolism, which enables it to sustain respiratory growth and efficiently utilize carbon sources even in the presence of excess glucose. To better understand and predict its metabolic behavior, genome-scale metabolic models are commonly employed. Within these models, Gene-Protein-Reaction (GPR) associations play a crucial role in the predictive performance of models. The previously developed iPichia genome-scale metabolic model for K. phaffii was further refined in this study through extensive manual curation of GPR associations, resulting in an updated model termed iPichia2. GPR associations were manually curated for 862 reactions based on the enzymatic functions and subcellular localizations of the associated proteins, using information obtained from the KEGG, NCBI, and UniProtKB databases. Gene essentiality analyses were subsequently performed using iPichia2, and the predicted results were compared with experimental data reported in the literature. This refinement of GPR associations improved the overall predictive performance of the model, with prediction accuracy increasing from 0.71 in iPichia to 0.727 in iPichia2. Similarly, the Matthews correlation coefficient (MCC) increased substantially from 0.1075 to 0.24, indicating improved agreement between predicted and experimentally observed gene essentiality phenotypes. Compared with iPichia, iPichia2 predicted fewer true positives (313 vs. 391), false positives (69 vs. 108), and false negatives (67 vs. 71), while the number of true negatives increased from 36 to 49. The higher accuracy and MCC obtained with iPichia2, despite its lower overall MEMOTE score (%34) compared with the original iPichia model (%52), highlights that model quality and predictive performance represent distinct aspects of genome-scale metabolic model evaluation. Overall, these results demonstrate that manual refinement of GPR associations can improve the predictive performance of GEMs without modifying their underlying reaction network. The improved GPR associations in iPichia2 are therefore expected to support more reliable identification of gene knockout targets and facilitate metabolic engineering and rational strain design in K. phaffii.