Metal-organic frameworks (MOFs) demonstrate complex behaviours under thermodynamic stimuli, often transitioning to non-crystalline states. These disordered states stand out due to their structural adaptability, e.g., the capacity to dissipate strain without significant material degradation. However, both experimental and computational challenges in analysing these highly disordered atomic structures left the transition behaviours of such MOFs largely unpredictable. Herein, I will address this gap by developing an in silico approach based on machine learning potentials to elucidate the mechanisms behind these crystalline-to-amorphous transitions. I will build a detailed understanding by examining how systematically varying the constituent building blocks in three archetypal MOF families impacts their transition behaviour. This requires me to overcome the traditional confines of crystalline modelling, and complement it by experimental validation, to enhance our collective insight into these phenomena. I will then rationalise this understanding via a coarse-grained model simplifying MOFs into building units with parameterisable degrees of internal flexibility and effective interactions. By determining how different material parameters influence the attainment of specific crystalline or non-crystalline states under external stimuli, this research will enable the design of MOFs with customisable responses to such triggers, opening new avenues in material science and applications.