Fig 1.
Structure of the work presented in this manuscript.
Left: computational prediction of PTM-related directed protein interactions (PTM-PPIs). A random forest model was constructed to predict PTM-PPIs based on hyperbolic topological features and network properties extracted from the hPIN. Right: quantitative proteomics data revealed dysregulated proteins in a cellular model of SCA1. The results were interpreted using predicted PTM-PPIs, providing directions for future experimental research on SCA1 pathogenesis.
Fig 2.
(A) Positions of nodes. Colors indicate node clusters. The circular plot at the center indicates the density of nodes in the angular dimension for all nodes (gray), and for the nodes of 295 directed PTM-PPIs with experimental evidence used for training a predictor: effectors (red) and targets (blue) (see text for details). (B) Top enriched GO BP term for each cluster.
Fig 3.
Properties of the 14 features that were used for building the model.
(A) Distributions of r, theta, degree centrality (DC), betweenness centrality (BC), closeness centrality (CC) and eigenvector centrality (EC), hyperbolic distance and r distance. (B) Feature importance values indicate the impact of each predictor on the prediction model. For this study, the angular coordinates of the of effector and target proteins in the hyperbolic space are the most highly significant predictors of the model, confirming the importance of the embedding of the hPIN in H2.
Fig 4.
Enrichment analysis of predicted effectors.
Left, negative predictions. Right, positive predictions. Colors from pink to grey indicate p-values from high to low; a lower p value suggests the proteins are more enriched in GO Biological Process and Molecular Function terms related to post-translational modifications. GO BP and MF terms are more enriched in the PTMs class, which supports the good performance of the model.
Fig 5.
Critical proteins associated with SCA1 pathology.
(A) A PPI network consisting of ataxin-1, its 13 PTM-PPIs and the 3 dysregulated kinases (n = 17). Node color corresponds to fold change (blue for down- and orange for upregulated proteins, respectively) and edge width to HIPPIE interaction score [24]. (B) Enrichment analysis for rare diseases highlighted an association of the critical proteins with SCA-related terms. (C) Known drugs which may restore the levels of at least 25% of the dysregulated proteins, ranked according to their combined score (p-value, adj. p-value, z-score).