Journal Papers A combinatorics based solution for input generation in the context of classification frameworks aimed to learning Interacting Protein Class


Abstract

Classification of proteins as Interacting - Non Interacting is a machine learning problem carried out either with informa- tive or with discriminative approaches by many researchers [21, 27, 9]. Missing information about the features describing the two classes is one of the main problems to deal with, and alternative models were proposed as possibly more effective solutions [12]. As a matter of fact, at the top positions in the list of desiderata for features aimed to Learning In- teracting Class there are: the universality (high coverage) and the discrimination power or class specificity. Here, the problem related to missing information in the features is bypassed by adapting single protein properties of universal coverage (i.e. available for all the yeast proteome) to become protein pairs attributes, still of universal coverage; the encoding information strategy is based on exploiting the apparent combinatorial nature of the association between single protein features to the classes of Interacting - Non Interacting protein pairs



Paper Details

Authors

M. Persico

Publication

Proceedings of the First ACM International Conference on Bioinformatics and Computational Biology

Download

/var/papers/1429006554-p684-persico.pdf

Language

English
.