Desciption of the server

Polarity is an essential feature of many cells, especially in differentiated, multicellular organisms. Many mammalian cell types exhibit a certain level of polarity, such as neurons, migratory cells, epithelial cells, and more. Epithelial cells possess a highly organized architecture establishing an apical-basolateral axis separated by tight junctions to maintain physiological barriers, as well as to deliver information to different regions of an organisms. Many times trafficking of these proteins from the Trans-Golgi Network to the plasma membrane does not occur in a single step, but rather via an indirect route through endosomal pathways, regulated by Short Linear Motifs (SLiMs).

Recently we constructed PolarProtDB, the most comprehensive collection of proteins localizing into polarized cells, with experimental evidence of their apical/basolateral localization. Based on this resource we collected predicted linear motifs that frequently occur in transmembrane protein localizing to apical/basolaretral domains and prepared fully connected Neural Networks predicing localization based on the distribution of these motifs. In addition, we also prepared Convolutional Neural Networks, recognizing patterns formed by adjacent amino acids. The final prediction takes into consideration both submethod.

PolarProtPred is the first prediction method predicting the localization of membrane proteins in epithelial cells, achieved reasonably high accuracy. 

 

Predictor	True Class	Balanced Accuracy	Sensitivity	Specificity	MCC	AUC
ApiBasoPred	Apical	0.89	0.87	0.91	0.78	0.96
PolarProtPred	Apical	0.81	0.85	0.78	0.61	0.93
	Basolateral	0.75	0.52	0.85	0.39	0.78
	Not apical/basolateral	0.68	0.62	0.73	0.35	0.81

PolarProtPred also displays prediction reliability, which highly correlates with the predictions accuracy.

Predictions are sorted according to their probability values, and then the accuracies and the lowest reliability measured on the subset of the benchmark set are plotted against their rank in the sorted list divided by the number of the proteins in the benchmark set (coverage). Left: apical/basolateral prediction, right: apical/basolateral/other prediction. Red: accuracy, blue: probability.