Read it again. The paper says that if you use a specific quantum mechanics model then you get results with several free parameters. These parameters can be tuned against 15 existing molecules and the equation gives some explanatory power of why the folding is non-Arrhenius.
Several things must occur before you can say it's real: does it apply to other molecules, what are the sources of the differences between prediction and measurement, do molecules while folding really undergo some sort of non-continuous motion and can it be detected and why hasn't it ever been seen before, and do the matching corresponding Newtonian+statistical dynamics give similar predictions?
Exactly! This high level model doesn’t draw on any quantum specific phenomena and is in many ways comparable to other stochastic dynamical models that have been developed to explain the non-Arrhenius behavior of protein folding using classical Langevin theory.
To argue for the inclusion of quantum theory in explaining protein dynamics is to assert that proteins undergo classically forbidden conformational transitions. The authors of the paper have provided no evidence for such transitions nor even hypothesized about how such transitions could arise. Their only argument against a classical description of protein folding is the inability of molecular dynamics simulations to model the folding-kinetics of some proteins. Seeing as molecular dynamics force fields only provide a rough and approximate description of the forces between atoms within a protein it should be no surprise these methods can’t accurately model the intricate conformational transitions in protein folding.
I am not sure Newtonian dynamics would give you similar predictions, but that doesn't mean the quantum predictions are wrong. However, with enough free parameters you can model anything. When you have 0 free parameters you have a real theory.
There are some extrange detals in the original paper.
For example the experimental points in the graphics do not have error bars. The x axis is 1/temperature, so the error should be small. But in the y axis I spect that the measures have a more visible error. Some of the graphics ("Protein A", "L9") have a very good theoretical fitting, so it is possible that the experimental data have very small errors. But in others graphics ("Engrailed Homeo Domain", "Trpcage(WT)") the points are scattered around the theoretical fitting, so perhaps the measures have a appreciable error or the theoretical fitting has problems.
And in the lasts graphics the fitting is not very good. In particular in "Villin Headpiece Subdomain" the fitting id horrble. I see that the experimental point lie in something like an "U" but the fit is like a shallow "^".
Another problem is that they have too many free parameters: "R" and "S" are free. And "Tc" is free too (it is an experimental value, not deduced from the model). But near the maximal point of the curve, everything is almost a parabola, so with 3 free parameters it is possible to fit almost everything. For example, for the first graphic ("Protein A") a possible fit using thee parameters is
ln(K_1)= -7.4161 * (1000/T - 3.1469)^2 + 12.418
They have a more difficult functional dependency, but with tree free parameters it is not clear that they have the correct model.
It seems obvious now, but I really would never have thought that quantum mechanics would matter here. I'm not sure that this simplifies anything. I wonder what D.E. Shaw research is going to do about it if this pans out? (They're the folks behind ANTON - see http://www.deshawresearch.com/)
Several things must occur before you can say it's real: does it apply to other molecules, what are the sources of the differences between prediction and measurement, do molecules while folding really undergo some sort of non-continuous motion and can it be detected and why hasn't it ever been seen before, and do the matching corresponding Newtonian+statistical dynamics give similar predictions?