Welcome to FuncPhos-STR

With the development of AlphaFold protein structure database, the structural characterization sheds new light on the long-standing question of functional relevance of phosphosites. Based on our proposed deep learning model named FuncPhos-SEQ, FuncPhos-STR is constructed herein with improved prediction for functional phosphosites. FuncPhos-STR consists of three feature encoding sub-networks (StrNet, SeqNet and SpNet) and a heterogeneous feature combination sub-network CoNet. Compared with FuncPhos-SEQ, the newly introduced StrNet is mainly used to extract structural and dynamics features related to phosphosite function based on the AlphaFold protein structure. SeqNet module implements the encoding of protein motifs to extract sequence context features. SpNet module employs graph embedding method to obtain PPI network topological features. Finally, CoNet module integrates the features from the three submodules, and then introduces SVM to score the function for input phosphosites, including general function, activity and molecular association.

FuncPhos-STR requires only the protein UniProt ID and concerned phosphosites as input. The residues with pLDDT < 50 would be deleted in the background preprocesses of loaded AlphaFold protein structure, and FuncPhos-SEQ is recommended for such situation. FuncPhos-STR is computationally efficient and pretty friendly for biologists without the knowledge of protein structures. The result page not only provides the predicted functional scores for input phosphosites, but also provides the interactive review in the context of protein 3D structures. Also, the structural and dynamics features used in the prediction are represented in line charts for better understanding the biophysical origins underlying the functional phosphosites.