Digital Subtraction Angiography (DSA) is a well-established imaging modality supporting treatment and diagnosis of vascular pathologies. Various clinical DSA acquisition protocols exist that provide qualitative blood flow information for vascular diseases. Velocity quantification algorithms primarily rely on tracking a contrast agent (CA) bolus through the vasculature. However, the true CA velocity is fast compared to the frame rate of angiographic images. Therefore, high blood velocities pose a challenge for these methods, as the bolus may flow through long vessel segments between two subsequent DSA frames. This problem can be mitigated by increasing the temporal resolution, or equivalently, the projection image frame rate. We propose a simulation-informed neural network approach to synthetically double the projection frame rate without additional patient dose. We evaluate the quality of the synthesized projection images and show the impact on bolus tracking algorithms. Synthesized projection images can be predicted with a mean absolute percentage error of \(2.5\pm 0.8\)  % in the inflow phase. Further, the synthesized projections qualitatively capture bolus dynamics more accurately compared to linear interpolation. Conceptually, our method allows extension to predicting multiple intermediate projection frames, which can be a valuable tool toward accurate quantitative vascular flow estimation.