Protein aggregation is associated with many neurodegenerative diseases. Understanding the aggregation mechanisms is a fundamental step in order to design rational drugs interfering with the toxic intermediates. This self-assembly process is however difficult to observe experimentally, which gives simulations an important role in resolving this problem. This study shows how we can proceed to gain knowledge about the first steps of aggregation. We first start by characterizing the free energy surface of the Abeta (16-22) dimer, a well-studied system numerically, using molecular dynamics simulations with OPEP coarse-grained force field. We then turn to the study of the NHVTLSQ peptide in 4-mers and 16-mers, extracting information on the onset of aggregation. In particular, the simulations indicate that the peptides are mostly random coil at room temperature, but can visit diverse amyloid-competent topologies, albeit with a low probability. The fact that the 16-mers constantly move from one structure to another is consistent with the long lag phase measured experimentally, but the rare critical steps leading to the rapid formation of amyloid fibrils still remain to be determined.