Integrative structure determination allows us to determine the structures of large macromolecular assemblies by combining data from complementary experimental methods with physical principles, statistical inference, and prior models. This talk will focus on efforts towards improving integrative modeling by my group in the last few years. Specifically, I will describe methods to optimize coarse-grained representations, annotate precision for integrative models, and obtain binding sites for intrinsically disordered regions. First, we developed methods to optimize the coarse-graining of subunits for an integrative modeling system, based on the amount of input information. The optimization is based on Bayesian model selection and we demonstrate that the optimized representations enable efficient sampling and result in models that fit the input data well. Second, we developed an efficient method to annotate regions of high- and low-precision in an integrative model ensemble. Low-precision regions can suggest where the next set of experimental data would be most impactful and high-precision regions can be used for identifying binding interfaces and suggesting new mutations. Finally, we developed a method to identify binding sites for intrinsically disordered regions (IDRs) of proteins in assemblies. Often, IDRs are challenging to localize in macromolecular assemblies, since they are associated with sparse structural as well as other interaction data. I will discuss a recent deep-learning method that we developed to predict binding regions for IDRs in large assemblies; the method outperforms AlphaFold3 on this task on benchmarks. Finally, I will also discuss emerging frontiers in the field: methods for utilizing data from cryo-electron tomography and methods that build upon AI-based structure predictions. In all, these methods will result in an improved spatio-temporal characterization of macormolecular structures in cells.