Poodle is a set of tools for the prediction of the secondary structures. The different programs uses different machine learning approaches.
|Author||Hirose S, Shimizu K, Kanai S, Kuroda Y, Noguchi T.|
|ML Method||two levels of SVMs|
Poodle-L is specialized on the prediction of long disordered regions (> 40 residues). It uses two SVMs. Each amino acid is represented by ten different physikochemical properties (10 dimensions). The first SVM predicts the probability of a 40-residue segment to be disordered. This output is used by the second SVM to predict predict the probability of a single residue to be disordered.
|Author||Shimizu K, Hirose S, Noguchi T.|
Poodle-S is specialized on the prediction of short disordered regions. This method defines seven regions on a protein. Depending of the physikochemical compositions and a PSI-Blast profile a SVM decides for one of the seven regions if it contains disorder.
The idea behind this is, that it is important where disorder happens and depending on the place the contribution to disorder of physikochemical features varies.
|Author||Shimizu K, Muraoka Y, Hirose S, Tomii K, Noguchi T.|
Poodle-W predicts which proteins are intrinsically disordered. It uses a method called Spectral Graph Transducer, which is a classification algorithm (see "Transductive learning via spectral graph partitioning" by Dr. Joachims and Poodle-W).
|Author||Hirose S, Shimizu K, Inoue N, Kanai S, Noguchi T.|
|Reference||CASP 8 Proceedings|
Poodle-I seems to predict the disorder of a protein by combining different tools in a workflow.