_PROBLEM CoEPrA-2006_Classification_001

_GROUP_NAME Joao Aires-de-Sousa

_GROUP_MEMBERS
Goncalo Carrera
Sunil Gupta
Yuri Binev
Joao Aires-de-Sousa

_ADDRESS
REQUIMTE, CQFB, Departamento de Quimica, Faculdade de Ciencias
e Tecnologia, Universidade Nova de Lisboa, 2829-516 Caparica,
Portugal. E-mail: jas@fct.unl.pt; Fax: +351 21 2948550

_MODELING_PROCEDURE
From the initial pool of 5787 descriptors given by the organizers, 433 
descriptors (file descriptors.txt) were selected to exclude:

- inter-correlation between descriptors above a Pearson correlation 
coefficient of 0.62.
- descriptors with values in the prediction set outside the range of the 
training set.

With the selected descriptors, classification models were built with 
Support Vector Machines. An ensemble of 5 SVMs was used for 
classification. The R program was employed (version 2.2.1  2005-12-20 
r36812) with the kernlab library (version 0.8-1). After manual 
optimization of parameters C, and nu, the values of C=8 and nu=0.008 
were chosen on the basis of 3-fold cross-validation results for the 
training set. The default value of epsilon (0.1) was used. A 
radial basis kernel "Gaussian" was used. The output files are named
class_svmx_output.txt (x from 1 to 5), and the parameters are in the 
file class_parameters_svm.txt.

Experiments with Random Forests were performed to select the most 
relevant descriptors to train SVMs. These experiments always resulted in 
higher 3-fold cross-validation errors than the experiments with the 433 
descriptors and no selection of features. This suggests a situation with 
different mechanisms of action involved, and made us decide to use the 
SVMs trained with 433 descriptors.

Alignment of the nona-peptides in the training set with the CLUSTALW 
program (multiple sequence alignment program for DNA or proteins) using 
default parameters at the EBI web site (http://www.ebi.ac.uk/clustalw/) 
produced a cladogram that was compared with the classification labels. 
The observation of a general good agreement between clusters of the 
cladogram and the class of the peptides, suggested that the 
classification of peptides in the prediction set could be assisted by the 
cladograms.

The final predictions for the peptides of the prediction set were 
obtained by majority vote of the 5 SVMs of the ensemble when the sum
of the 5 probabilities was more than 3.5. When the sum of the 5 
probabilities was less than 3.5, the peptide was predicted from the class 
of its neighbors in the cladogram built with the sequences of peptides in 
the training and prediction sets together (file cladogram-clustalw_tr_te.jpg).
For that reason, the predictions obtained from the ensemble of SVMs were 
changed for the peptides 16, 28, 30, 41, 44, 53, 62 and 75.


_PREDICTION
Obj_00001 -1
Obj_00002 +1
Obj_00003 -1
Obj_00004 -1
Obj_00005 +1
Obj_00006 +1
Obj_00007 -1
Obj_00008 -1
Obj_00009 -1
Obj_00010 +1
Obj_00011 +1
Obj_00012 +1
Obj_00013 +1
Obj_00014 +1
Obj_00015 -1
Obj_00016 -1
Obj_00017 -1
Obj_00018 +1
Obj_00019 -1
Obj_00020 -1
Obj_00021 +1
Obj_00022 +1
Obj_00023 -1
Obj_00024 +1
Obj_00025 +1
Obj_00026 +1
Obj_00027 +1
Obj_00028 +1
Obj_00029 +1
Obj_00030 +1
Obj_00031 +1
Obj_00032 +1
Obj_00033 -1
Obj_00034 -1
Obj_00035 +1
Obj_00036 -1
Obj_00037 -1
Obj_00038 -1
Obj_00039 -1
Obj_00040 +1
Obj_00041 +1
Obj_00042 -1
Obj_00043 -1
Obj_00044 +1
Obj_00045 +1
Obj_00046 -1
Obj_00047 +1
Obj_00048 +1
Obj_00049 +1
Obj_00050 +1
Obj_00051 +1
Obj_00052 +1
Obj_00053 -1
Obj_00054 -1
Obj_00055 +1
Obj_00056 +1
Obj_00057 +1
Obj_00058 -1
Obj_00059 +1
Obj_00060 -1
Obj_00061 -1
Obj_00062 +1
Obj_00063 +1
Obj_00064 -1
Obj_00065 +1
Obj_00066 +1
Obj_00067 +1
Obj_00068 -1
Obj_00069 +1
Obj_00070 -1
Obj_00071 -1
Obj_00072 +1
Obj_00073 -1
Obj_00074 -1
Obj_00075 -1
Obj_00076 -1
Obj_00077 -1
Obj_00078 +1
Obj_00079 -1
Obj_00080 +1
Obj_00081 +1
Obj_00082 -1
Obj_00083 -1
Obj_00084 +1
Obj_00085 -1
Obj_00086 -1
Obj_00087 +1
Obj_00088 -1