Protein Structure and Function. Application of Bioinformatics Methods - John Rigden 2014

Integrated Servers for Structure-Based Function Prediction
ProKnow
Combining Predictions

When all processes are completed, the Functions (i.e., GO terms) associated with recurrently identified features are combined using Bayes' theorem to assign weights, which allows for The Significance of each predicted term to be assessed. Only terms related to molecular function and biological processes are considered; terms pertaining to intracellular localization are excluded. The significance of each predicted term is represented by three numbers. The first is the Bayesian weight corresponding to the probability—ranging from zero to one—that the term prediction is correct. The second number is the feature rank, which indicates how reliable a particular GO assignment is considered to be for ranking first. This is due to the diverse origins of such Assignments: they may be curated, obtained through direct observation, inferred from structural or sequence similarity, and so on. Consequently, their reliability varies, being highest for assignments supported by direct experimental evidence. The source of the annotation is indicated as a feature code in the GO data. The ProKnow server translates each feature code into a rank for numerical evaluation of its reliability, and averaging the ranks of multiple predictions yields the feature rank. The third significance indicator is the number of keys, which equals the number of weights used to calculate the Bayesian weight, reflecting the number of ProKnow server Methods that contributed to the given GO term prediction.