5.2  The quality of predictions by the current SABR-P algorithm and future work
The current SABR-P predictive algorithm is naïve and it is not expected that it will resemble closely the final refined form of the algorithm, which will based on more rigorously on principles closer to those of the GOR method [20], the Hyperbolic Dirac Net [[21], [22], [23]], the association Q-UEL language [24], and the BionIngine implementation including its new algorithms [[25], [26], [27], [28]]. The impression of good performance for the current SABR-P method largely arises from the fact that it is only required to predict the sialic acid glycan binding properties of whole domains or proteins, not highly localized subsequences or surface patches. In essence, the method is really doing little more than capture and quantify in an algorithm the visual inspection of sugar binding domains and proteins and the observations of other workers as discussed above. However, the method was only required to help explore potential non-covalent sialic acid glycan binding sites in the spike glycoprotein, and in that regard it has proven adequate and valuable for present purposes. It also suggests a more refined approach may perform well because false positives and false negatives were mainly just over the boundary and just under it respectively. Resolution should be increased.