In any case, one cannot ignore the fact that the characteristics of the protein environment also can play an important role by being able to modify protein structures and, consequently, interfaces. Additionally, it is also clear that if a protein is interacting with two or more different partners, different interfaces may be formed for each partner. A careful literature review will quickly confirm that although there are several recently published studies regarding the characteristics that could determine differences between interface-forming residues and free surface residues, there is no general agreement about exactly how proteins associate with each other and which descriptors of their characteristics are suitable for elucidating this mechanism. Also, by comparing the interface area against the rest of the free surface is a common procedure during attempts to characterize the main differences between those two classes. This type of comparison has been described in recent studies, including those cited above Fulvestrant. A variety of models and descriptors were explored to build protein-protein interface classifiers. Promate and PINUP used linear scoring functions, while PPI-Pred used a support vector machine approach, SPPIDER and cons-PPISP used a neural network model, and Meta-PPISP combined the results of cons-PPISP, Promate and PINUP as a meta-predictor. In contrast to the method proposed in this study, the six mentioned models make use of amino acid sequence conservation and propensity. How important is this difference among StingLDA and other mentioned algorithms could only be accessed adequately if proper analysis is done on how often the conservation property could not be used in known protein universe. It is known that structural genome projects used high-throughput techniques to produce and then deposit in the PDB thousands of new structures. For instance, half of the protein structures solved during the year of 2005 came from structural genome initiatives, including structures of the so-called orphan proteins. Orphan proteins are organism-specific proteins, i.e. GANT61, they have no homologue protein in other lineages. Estimates are that up to one third of the genes/ proteins from whole known genomes accounts for orphan proteins. Ekman et al. show, using the structural classification of protein, that up to 25% of the known non-redundant protein structures from bacteria are from orphan proteins or from proteins having an orphan domain. Also, up to 21% of known protein structures in Eukarya kingdom and 24% in Archaea kingdom follow the same trend. Operating in such scenario where limitations imposed by orphan protein existence restricts the use of aforementioned algorithms dependent on conservation parameter for predicting interface residues, would clearly lead to unreliable results. Therefore, the strong demand is created for the development of more general approaches for IFR prediction which would have similar performance to conservation dependent algorithms, yet without the use of evolutionary-related attributes for prediction. The Sting-LDA was produced having in mind this demand as well. We report results on the classification of the 20 naturally occurring amino acids into two distinct classes: IFR and FSR, by using several amino acid descriptors from the BlueStar STING database. BlueStar STING has been used previously for predicting enzyme class, protein-ligand analysis, protein mutant analysis, and protein-protein interaction pattern analysis, mostly because BlueStar STING offers easy access to a very rich repository of protein characteristics.