KAT-Net: a hyperbolic routing framework for joint lysine acylation site prediction with shared sequence-structure representations.

KAT-Net achieves superior joint lysine acylation site prediction by integrating sequence and structural data with hyperbolic routing, outperforming single-task models in benchmark tests.

• Why it matters: Accurately predicting lysine acylation sites is crucial for understanding protein regulation, but current methods lack joint modeling of different acylations and structural context, limiting biological insight.
• What they did: The authors developed KAT-Net, a multi-task framework that combines ESM-2 representations, structural modeling via AlphaFold2, and hyperbolic routing, trained on non-histone acetylation data with 1,000+ samples.
• The result: KAT-Net demonstrates favorable performance in benchmark and external evaluations, with analyses revealing modification-specific positional and structural patterns, supporting its utility for lysine-acylation prediction and candidate-site prioritization.