The Euler-Lagrange approach, based on Direct Numerical Simulation (DNS) and Large-Eddy Simulation (LES) for the fluid, is applied to particle-laden turbulent flow in a channel. Explicit subgrid modeling of the turbulent stresses is adopted, while the particle motion includes small turbulent scales based on approximate deconvolution of the LES field. Results for turbulent flow in a channel at Reτ = 150 are discussed, focusing on one-way coupled point-particle statistics at three Stokes numbers. DNS provides a point of reference for assessing LES with different sub-filter eddy-viscosity models: Smagorinsky, Van Driest- Smagorinsky and the dynamic model are studied. Clustering and segregation of particles near the wall, due to turbophoresis, is strongly related to the quality of the LES velocity field and the approximate reconstruction of the smaller resolved scales. It is shown that deconvolution up to second order allows to better describe the particle statistics near a solid wall; deconvolution at higher order yields rather marginal additional improvements.