Hello everyone!

A few years ago i have developed an algorithm to combine the results from multiple classifiers (ex: signal peptide predictors, beta-barrel predictors) in a consensus by using an unsupervised approach. Briefly, the algorithms receives an matrix with the predictions generated by "n" classifiers for "m" proteins and then defines a weight for each classifier based on how much it's results were "confirmed" by the others.

It was developed in the context of a reverse vaccinology study, where we had to run multiple predictors for the same protein property and then combine them to rank the proteins that were more likely to be good vaccine targets. As no good reference database was available for some properties, we decided to create an unsupervised method.

The code we used on the paper was refactored and now can be installed from the Python Package Index (PyPI). https://github.com/fredericokremer/covira

Can anyone give me some feedback about how to improve the method, or it's implementation?

More Frederico Schmitt Kremer's questions See All
Similar questions and discussions