The server implements a method for predicting local structure by amino acid sequence for the most popular set of structural prototypes of 16 PB with a length of 5 amino acids [1].
That is, prediction determines to which of the 16 structural classes each pentapeptide of an amino acid sequence belongs to.
Prediction is based on physicochemical properties [2] and RMSD-based statistics. The prediction method does not in any way use information about homologous proteins.
There are two server operation modes:
"Predict Local Structure Classes" - actually predicting structural classes by sequence. The amino acid sequence is fed to the input.
"Compare prediction for Known Structure" is a prediction quality analysis for a known protein chain structure. A PDB file is input. You must select a PBD file and do not forget to select the ID of chain

[1] De Brevern A, Etchebest C, Hazout S. Bayesian probabilistic approach for 275 predicting backbone structures in terms of protein blocks. Proteins: Structure, 276 Function, and Bioinformatics. 2000;41(3):271–287.

[2] Kawashima S, Pokarowski P, Pokarowska M, Kolinski A, Katayama T, Kanehisa 304 M. AAindex: amino acid index database, progress report 2008. Nucleic acids 305 research. 2007;36(suppl 1):D202–D205. 306