On the limits of marginal protein folding stability: methodological goals and drawbacks

Why do proteins remain at the edge of stability rather than evolving toward maximal thermodynamic robustness? Can we trust the computational methods that inform about when and how stability falls off that edge? Marginal stability, the confinement of protein folding free energies to a window of approximately 5-15 kcal/mol, is broadly observed in biophysics, and its mechanistic causes, involving evolutionary and biophysical properties, remain under debate and require accurate predictions. In this regard, protein folding stability is currently predicted using both traditional structure-based homology-modeling methods and modern sequence-based deep-learning methods. However, comparisons among these methodologies remain scarce.

Using four viral proteins of biomedical interest and considering that rapidly evolving viral proteins provide a useful system for understanding thermodynamic tolerance under the competing demands of structural integrity and adaptive flexibility, we compared predicted protein stability from both approaches across independent random mutational trajectories and naturally occurring sequence variants available in public databases. The results showed a general methodological impasse. As random mutations accumulated, structure-based methods predicted a progressive decline in stability toward collapse, a pattern driven by the enforced conservation of protein structure. In contrast, sequence-based methods failed as sequences moved into uncharted regions of parameter space. Together, protein stability landscapes predicted using current computational approaches are inherently shaped by technical limitations, leading to uncertainty in the evaluation of marginal stability and stability changes associated with biological processes.