Using microenvironments to identify allosteric binding sites

Protein amino acid residues can be classified by their chemical properties and data mining can be used to make predictions about their structure and function. However, the properties of the surrounding residues contribute to the overall chemical context. This paper defines microenvironments as the spherical volume around a point in space and uses these volumes to determine average properties of the encompassed residues. The approach to index generation rapidly constructs microenvironment data. The averaged chemical properties are then employed in allosteric site prediction using support vector machines and neural networks. The results show that index generation performs best when microenvironment radius matches the granularity of the index and that microenvironments improve the classification accuracy.