Knowledge-based Fragment Binding Prediction
Figure 2
FragFEATURE predicts fragments for a protein pocket of interest.
Given a pocket of interest as a series of microenvironments (semi-transparent circles), we compare each microenvironment to knowledge base microenvironments of the same type to retrieve the five most similar non-homologous neighbors. Each neighbor has a list of bound fragments for which a hypergeometric p-value is determined. For spatially proximal microenvironments (orange, blue, and magenta circles), we combine fragment hypergeometric p-values for shared fragments to generate Fisher's p-values. Denoted with an asterisk are statistically significant fragments with p-value(**)<p-value(*).