A Lagrangian-Eulerian based, large-eddy simulation (LES) study of particle-laden turbulent flows is presented. The main focus of the study is a sub-grid two-phase interaction model based on an approximate deconvolution method. This model contributes as a particle source term to the gas phase in the transport equation of sub-grid kinetic energy which, in turn, is used for modeling the sub-grid shear stress. The performance of the model and the overall ability of LES in predicting particle/droplet-laden turbulent flows is discussed based on the test flow configurations of a particle-laden gas jet, particle-laden isotropic decaying flow, and high-speed diesel sprays. The mean and turbulent flow statistics from LES are shown to compare very well with previous experimental and DNS studies. The dependence of the two-phase turbulence interaction model on particle loading and Stokes number is found to be consistent with previous DNS analyses. For the isotropic cases, the LES results correctly capture the effect of the particles on different length scales of the flow. The characteristics of different mechanisms contributing to sub-grid kinetic energy are found to be consistent with the previous DNS results. The LES results are shown to be in good agreement with the near-nozzle diesel spray measurement for quantities such as the transverse integrated mass, the mass-averaged axial spray velocity, and the spray momentum. The particle contribution to sub-grid kinetic energy is found to be significant in the conditions of high-particle Stokes numbers, e.g., in the high-speed diesel sprays.