Computational fluid dynamic (CFD) simulation is a leading approach to accurately model state of physical environment, e.g. blood pressure in cardiovascular system. However, due to the high level of detail of simulated system that CFD methods require they tend to be computationally expensive creating a need for ml-assisted solutions. While methods incorporating deep learning for pressure drop and fractional flow reserve (FFR) estimation are advancing rapidly, they suffer from need for large amount of real data, which is scarcely available, simulated under many possible scenarios presenting a challenge from computational point of view. In this work we present Hemodynamic Feature Regression (HFR)–method for pressure drop estimation that by incorporating hemodynamical modeling is able generalize from synthetic to real samples. We test the pressure drop predictions on real coronary geometries compared to CFD simulations.