The development of computational efficient models is essential to obtain a detailed characterization of the mechanisms underlying the folding of proteins and the formation of amyloid fibrils. Structure-based computational models (Go-model) with C alpha or all-atom resolutions have been able to successfully delineate the mechanisms of folding of several globular proteins and offer an interesting alternative to computationally intensive simulations with explicit solvent description. Here, we explore the limits of Go-model predictions by analyzing the folding of the nonglobular repeat domain proteins Notch Ankyrin and p16(INK) and the formation of human islet amyloid polypeptide (hIAPP) fibrils. Folding trajectories of the repeat domain proteins revealed that an all-atom resolution is required to capture the folding pathways and cooperativity reported in experimental studies. The all atom Go-model was also successful in predicting the free-energy landscape of hIAPP fibrillation, suggesting a "dock and lock" mechanism of fibril elongation. We finally explored how mutations can affect the co-assembly of hIAPP fibrils by simulating a heterogeneous system composed of wild-type and mutated hIAPP peptides. Overall, this study shows that all-atom Go-model-based simulations have the potential of discerning the effects of mutations and post-translational modifications in protein folding and association and may help in resolving the dichotomy between experimental and theoretical studies on protein folding and amyloid fibrillation.