Crystallizing proteins is the bottleneck to systematically determining their structures. Identifying the solution conditions under which these macromolecules crystallize should be equivalent to determining their phase diagram, yet one typically resorts to combinatorial rather than physics-guided sampling to tackle this difficult problem. Although several soft matter-based ''patchy particle'' models have been invoked to rationalize protein assembly, the interactions that drive protein crystallization are insufficiently characterized for these models to be of much practical use. As a proof of concept of their relevance we simulate solvated proteins to parameterize patchy models, and find the self-assembly phase diagram to agree with crystallization experiments. By generalizing these models, we explore the range of conditions over which crystal assembly is possible, providing guidelines for crystallizing some more recalcitrant proteins.