5.4  Future challenges
Specifically, a larger sample may require a threshold adjustment or corresponding rescaling, perhaps resulting in a deterioration of performance particularly in regard to distinguishing lectins. Nonetheless, it is noteworthy that ultimately human glycan binding proteins have to overcome the same problem as the above kind of prediction algorithm. While this broad range of sugar recognition by the mannose-binding lection permits that lectin to interact with a wide selection of pathogens (viruses, bacteria, yeasts, fungi and protozoa) decorated with such sugars, there must be some kind of distinguishing aspect such that is not decoyed by the sialic acid glycans of the human host. A more mundane problem in extending the study is that the correct state as sialic acid glycan binding, other sugar binding, or not binding any kind of sugar, may be uncertain or a matter of degree. Further studies at time of writing suggested only about 70% for each of accuracy, sensitivity, and specificity, but this larger set is, as yet, of dubious quality for the purpose. Some proteins were believed, rather than known, to bind sialic acid glycans, binding might be weak or less specific or of multiple types, or the domain or approximate location of the binding site can be unclear. Related to that is a difficulty that the performance of any prediction method of this kind is defensible, and possibly unfairly defensible, in regard to false positives: it may be that experiment shows that a particular virus predicted to bind sialic acid glycans does not specifically do so, but perhaps it once did, in evolutionary terms. This is particularly relevant in regard to studying coronaviruses because, as discussed above, many coronaviruses certainly do bind sialic acid glycans. Of course, the prediction method would then still be subject to the criticism that it insufficiently sophisticated to manage the impact of small changes. For purely theoretical methods, that may be an issue for some time: in the present author's experience even simulations of binding of sugars to proteins in atomic detail tend to be difficult in view of the complex role of water molecules. For example, water molecules commonly represent protein-to-sugar bridges as discussed in this paper.