Multiscale integrative modeling stands at the intersection between experimental and computational techniques to predict the atomistic structures of important macromolecules. In the integrative modeling process, the experimental information is often integrated with energy potential and macromolecular substructures in order to derive realistic structural models. This heterogeneous information is often combined into a global objective function that quantifies the quality of the structural models and that is minimized through optimization. In order to balance the contribution of the relative terms concurring to the global function, weight constants are assigned to each term through a computationally demanding process. In order to alleviate this common issue, we suggest to switch from the traditional paradigm of using a single unconstrained global objective function to a constrained optimization scheme. The work presented in this thesis describes the different applications and methods associated with the development of a general constrained optimization protocol for multiscale integrative modeling. The initial implementation concerned the prediction of symmetric macromolecular assemblies throught the incorporation of a recent efficient constrained optimizer nicknamed mViE (memetic Viability Evolution) to our integrative modeling protocol power (parallel optimization workbench to enhance resolution). We tested this new approach through rigorous comparisons against other state-of-the-art integrative modeling methods on a benchmark set of solved symmetric macromolecular assemblies. In this process, we validated the robustness of the constrained optimization method by obtaining native-like structural models. This constrained optimization protocol was then applied to predict the structure of the elusive human Huntingtin protein. Due to the fact that little structural information was available when the project was initiated, we integrated information from secondary structure prediction and low-resolution experiments, in the form of cryo-electron microscopy maps and crosslinking mass spectrometry data, in order to derive a structural model of Huntingtin. The structure resulting from such integrative modeling approach was used to derive dynamic information about Huntingtin protein. At a finer level of resolution, the constrained optimization protocol was then applied to dock small molecules inside the binding site of protein targets. We converted the classical molecular docking problem from an unconstrained single objective optimization to a constrained one by extracting local and global constraints from pre-computed energy grids. The new approach was tested and validated on standard ligand-receptor benchmark sets widely used by the molecular docking community, and showed comparable results to state-of-the-art molecular docking programs. Altogether, the work presented in this thesis proposed improvements in the field of multiscale integrative modeling which are reflected both in the quality of the models returned by the new constrained optimization protocol and in the simpler way of treating the uncorrelated terms concurring to the global scoring scheme to estimate the quality of the models.