Protein based therapeutics have emerged as a successful class of pharmaceutical. However, it is well known that most of the current therapeutic protein discovery and development processes are not particularly productive. Efficient production of protein is required to meet the growing demands and increasing expectation of the patients and healthcare providers. The major obstacles during biopharmaceutical production are protein expression level and aggregation which may cause serious problems. Computational tools to predict protein solubility and aggregation propensity are therefore of great interest to biopharmaceutical development. In this study, a computational tool for prediction of protein soluble expression based on protein structure and surface electrostatic properties is experimentally validated with human erythropoietin (HuEPO) as the model protein. A significant inverse correlation between positive surface patches and HuEPO expressibility was found in mammalian cells, probably by influencing the translation efficiency, protein stability or secretion. The results demonstrated the potential of the predictive computational algorithm as a design tool in rational protein engineering to improve the protein developability. In summary, optimization of molecular patches on the surface of proteins may be a viable strategy for enhancing protein soluble expression and therefore a possible solution for difficult-to-express proteins.