Fig. 1

This work is logically divided into three phases: dataset extraction (in blue), model training (in orange), and sensitivity prediction (in yellow). In the first phase, we performed numerical simulations on a set of manually curated BPs. This process resulted in a preliminary dataset (DyBP) where BPs were annotated with sensitivity information. Then, we mapped the sensitivity relationships in the DyBP dataset to a PPIN, producing a second dataset (DyPPIN). Both datasets are made publicly available. In the second phase, we trained a DGN on the DyPPIN dataset to predict sensitivity relationships, assessing its performances in different use cases. Ultimately (third phase), the end user can use the trained DGN to predict the sensitivity across different portions of the PPIN. The inference phase is visually detailed in Fig. 4