Because they develop on a wide range of length and timescales, structural and dynamical properties of macromolecular systems are difficult to study by atomistic computer simulations. To overcome these limitations, many descriptions of polymeric liquids with reduced internal degrees of freedom, or coarse-grained (CG) descriptions, have been proposed. This seminar is an overview of some unresolved questions in the field of coarse-graining and multiscale modeling of polymeric liquids (melts) and of our attempts to address them. CG models are derived by averaging out local degrees of freedom at a lengthscale shorter than a characteristic length, and obtaining in this way a new description in the coordinates of the resulting CG units, where the system interacts through an effective potential. The new CG description has the advantage that the length and timescale of the elementary step in the CG simulation are increased, the CG simulation accelerated, and the maximum length and timescales that the CG simulation can cover significantly extended. CG methods also have the disadvantages with respect to atomistic simulations of losing information of the local scale, being thermodynamically inconsistent, and producing unrealistically fast dynamics. We have recently proposed a coarse-grained method, based on liquid state theory, where polymers are represented as chains of soft spheres. The simplicity of the model allows for analytical solutions of many of the relevant properties, both static and dynamic, and shows thermodynamic consistency for the CG model. A first-principle analytical procedure to reconstruct atomistic dynamics from CG simulations is presented and tested against experiments and atomistic simulations of polymer melts of different chemical structures, showing good agreement.