Molecular docking (MD) simulation is one of the four steps of the rational drug design. By the use of a fully-flexible receptor (FFR) model, protein flexibility can be explicitly considered during the drug design process. FFR models are composed of hundreds to several thousands of conformations to simulate the receptor flexibility in cell environments. So, for each conformation in the FFR model a MD simulation is executed and analyzed against a small molecule. However, it presents an important challenge due to small molecules databases, e.g. ZINC, have more than 21 million compounds available. It is computationally very demanding task to perform virtual screening of millions of ligands using an FFR model in a sequential mode. This paper introduces a data-flow pattern to perform massivelyparallel MD simulations using FFR models, named Self-adaptive Multiple Instances (P-SaMI). Based on a previously-clustered FFR model, P-SaMI permits to selectively perform experiments by the use of the Free Energy of Binding (FEB) output, used as quality criterion: smaller FEBs mean better results. The main goal achieved on using P-SaMI was the significant reduction of the number of docking experiments comparing with exhaustive ones, for a specific small molecule.