Enhancing k<sub>cat</sub> prediction through residue-aware attention mechanism and pre-trained representations.

Permutation tests confirmed that the differences between attention-derived residues and random selection were highly significant for the top 1–12% thresholds ( p = 0.000999), moderately significant at 13% ( p = 0.00599) and 14% ( p = 0.01798), but became statistically insignificant beyond the 15% threshold ( p > 0.05).

also in 2,182 other papers