Penrose developed a theory called "twistor theory" that tries to describe the universe using twistors. (https://en.wikipedia.org/wiki/Twistor_theory)
From what I've read, there are a lot of papers researching on twistors and applying them to numerous processes and physical models/theories.
In this entry (https://motls.blogspot.com/2017/02/a-story-about-roger-penrose.html) in Lubos Motl's blog that says "It's not quite clear whether twistors are totally sufficient to describe quantum gravitational phenomena in D=4" and in the spanish wikipedia entry of twistor theory it says "For a time it was hoped that the theory of twistors constituted by itself a direct path to quantum gravity, but this, at present, is considered unlikely".
Also, I found that neural networks could not be properly described by them (but the book itself where I read that said that some modification of it could do it) and renormalization group calculations as it is indicated here (Can modern twistor methods to calculate scattering amplitudes be applied to renormalization group calculations?) but the comments and the answers say that while being not the best option (it makes mathematics difficult when trying to apply twistors there) it is not impossible.
So how can twistor theory overcome these problems? If the original form of it cannot overcome these, would a modified version of twistor theory or a generalization of it (as in the case of neural networks I said before) do the job? (From what I've read it is usually said that twistors themselves have these problems, so maybe modifying or adding something to twistor theory or making a generalization of it could solve these) Is it there some specific example of what I'm looking for?
And if twistor theory in any form could do this, what about (also) Penrose's spin networks? (https://en.wikipedia.org/wiki/Spin_network)