The non-linear and multiscale nature of turbulent flows is further complicated in the presence of inertial particles. Intimate coupling between the phases may lead to a high degree of spatial segregation that reorganizes the structure of the underlying turbulence. The wide range of relevant length and timescales associated with fluidparticle systems poses significant challenges in understanding and predicting their behavior. In recent years, the advent of petascale computing has enabled the direct numerical simulation (DNS) of large-scale turbulent flows, though DNS of particle-laden flows remains severely limited. This work presents methods to alleviate previous numerical constraints on the computational grid when considering finite-size particles. Volume filtered equations for the carrier phase are derived in detail for variable-density flows in the presence of particles and solved in a highly-scalable Eulerian-Lagrangian framework. The filter introduces a separation in length-scales during the interphase exchange process, where everything smaller than the support of the filtering kernel requires modeling (e.g., surface reactions and drag), and everything larger than the support of the filtering kernel is captured explicitly. To remain computationally tractable, the filtering procedure is solved in two steps, by first transferring the particle information to the nearest neighboring cells, and then making use of an implicit diffusion operation. In flows that exhibit strong spatial segregation in particle concentration, a separation of length scales must be established when extracting Lagrangian statistics. To accomplish this, an adaptive spatial filter is employed on the particle data with an averaging volume that varies with the local particle-phase volume fraction. The filtered Euler-Lagrange formalism is shown to yield highly accurate and physical results for large-scale particle-laden flows from the dilute to dense regime. An analysis of chemically reacting species in circulating fluidized bed risers reveals that the non-homogeneities caused by the formation of clusters significantly reduces the efficiency of the conversion process. To better understand the fundamental nature of particle clustering and its effects on the carrier-phase turbulence, a canonical flow is introduced, referred to as cluster-induced turbulence (CIT). Simulations of fully-developed, gravity-driven CIT are investigated, revealing for the first time the local instantaneous distribution of particle-phase dynamics in collisional gas-solid flows.