Читать книгу Bioinformatics and Medical Applications - Группа авторов - Страница 75

There is a practical interest, besides the scientific interest, in aiming efforts to determine the predominant function of a protein. Let us start by saying that proteins do not have a single action on a pathogen; on the contrary, all proteins have action on several pathogens. For this reason, when examining a database specialized in proteins, it is common to find that the same protein is reported several times, with a different predisposition.

On the other hand, determining the action or function of a protein involves costly experiments and/or clinical trials, without mentioning that there are proteins with pathogenic action that are increasingly difficult to find in nature; as in the case of SCAAPs (selective cationic amphipathic antibacterial peptides) that are highly toxic to bacterial membrane and harmless to erythrocytes.

SCAAPs are also very short 6aa proteins – 14aa; for all these characteristics, they are very valuable in the production of pharmaceutical drugs; however, it is increasingly difficult to find them in nature.

In this scenario, it is very useful to have computational mathematical predictions that can identify the preponderant function of a protein only by taking its sequence. This enables the inspection of databases to search peptides with a specific function. Of course, this will not prevent experimental testing, but it will substantively reduce the proteins tested.

There are several types of classification for prediction algorithms of the protein predominant function, such as the representation of proteins in three-dimensional space rather than in linear representation. Others use stochastic algorithms instead of deterministic algorithms that may or may not evaluate physico-chemical properties.

In this work, we used two main divisions: the supervised algorithms and the non-supervised algorithms, both classifications are discussed below.