Prior probability is 1/3 for the first step; equiprobable.
p(F|feature-clue) = p(F) * p(feature-clue|F) / Z
p(F|feature-clue) = The posterior probability or the updated probability
p(F) = The prior probability
p(feature-clue|F) = The likelihood of function F given a clue from protein feature
Z = Normalization constant, Summation [ p(feature-clue|F) * p(F) ]
We assume that if we analyze the protein, we would be able to find certain descriptors in proteins called "features" which can give us clues to a set of functions. These features could be the sequence of the protein, the fold, motifs or functional linkages etc. An estimate of the strength of the clue to a function is given by an evidence value (essentially a weight). All such evidences are weighted using the Bayesian theorem to arrive at an updated weight for a function. The final result is a set of function and weights corresponding to the evidence found in the structure.
| Clue 1 | Clue 2 | Clue 3 | |||
| Weight | Function | Weight | Function | Weight | Function |
| 0.45 | Function 1 | 0.50 | Function 1 | 0.33 | Function 1 |
| 0.45 | Function 2 | 0.19 | Function 2 | 0.33 | Function 2 |
| 0.10 | Function 3 | 0.31 | Function 3 | 0.33 | Function 3 |