The self-assembly of proteins into protein quaternary structures is of fundamental importance to many biological processes, and protein misassembly is responsible for a wide range of proteopathic diseases. In recent years, abstract lattice models of protein self-assembly have been used to simulate the evolution and assembly of protein quaternary structure, and to provide a tractable way to study the genotype-phenotype map of such systems. Here we generalize these models by representing the interfaces as mutable binary strings. This simple change enables us to model the evolution of interface strengths, interface symmetry, and deterministic assembly pathways. Using the generalized model we are able to reproduce two important results established for real protein complexes: The first is that protein assembly pathways are under evolutionary selection to minimize misassembly. The second is that the assembly pathway of a complex mirrors its evolutionary history, and that both can be derived from the relative strengths of interfaces. These results demonstrate that the generalized lattice model offers a powerful new framework for the study of protein self-assembly processes and their evolution.Protein complexes assemble by joining individual proteins together through interacting binding sites. Because of the long time scales of biological evolution, it can be difficult to reconstruct how these interactions change over time. We use simplified representations of proteins to simulate the evolution of these complexes on a computer. In some cases the order in which the complex assembles is crucial. We show that biological evolution increases the strength of interactions that must occur earlier, and decreases the strength of later interactions. Similar knowledge of interactions being preferred to be stronger or weaker can also help to predict the evolutionary ancestry of a complex. While these simulations are not realistic enough to make exact predictions, this general link between ordered pathways in assembly and evolution matches well-established observations that have been made in real protein complexes. This means that our model provides a powerful framework for the study of protein complex assembly and evolution. Introduction 1 Many proteins self-assemble into protein quaternary structures, which fulfill a multitude 2 of functions across a wide range of biological processes [1]. Abstract models trade off 3 the complexity arising from conformations, buried surfaces, cooperative binding, etc., 4 but still retain qualitative realism. A general class of polyomino tile self-assembly 5 models have strong analytic potential while maintaining semblance to protein quatenary 6 structure. 7 The polyomino self-assembly model [2] combines lattice tile self-assembly with a 8 quantification of biological complexity, examining the relationship between genetic 9 description length and phenotypic complexity. The same model was developed and 10 expanded with evolutionary dynamics by Johnston et al. [3], and used to probe general 11 properti...