In Russian GOST R34.11-2012 LPS-transformation is used. LPS gets all of its non-linearity from the 8-bit S-box S, which apparently has been designed to offer resistance against classical methods of cryptanalysis. Its differential bound is P = $8/256$ and best linear approximation holds with P = $28/128$ . There seem to be no exploitable algebraic weaknesses.
Can you explain what is "differential bound of the S-Box"? I think that it's a probability: P(S($x_1$) $\oplus$ S($x_2$) = b | $x_1$ $\oplus$ $x_2$ = a) for random $a, b$. Is it a correct definition? And LPS seems to be no exploitable algebraic weaknesses because there are 64 S-Boxes and so, differential probability of the LPS at all is $(8/256)^{64}$?