Network motif-based identification of transcription factor-target gene relationships by integrating multi-source biological data

Table 2 Performance of SVM classifiers evaluated via LOOCV.

Input data	Auto regulation	Feed-forward loop	Single input	Multiple input	Average error
Gene expression data	3.7	48.1	17.6	24.8	23.6
Gene expression data and binding site sequence information	3.5	30.3	10.8	18.6	15.8

To evaluate the performance of the SVM classifiers, LOOCV was performed. To examine whether both gene expression data and binding site sequence information are needed in classifying TFs into different NM categories, we built SVM classifiers using only gene expression data. If only gene expression data are considered as input data, the average test error is 23.6%. After incorporating binding sequence data into the input data, test error has been reduced 15.8%. The increased performance implies that the encoded binding site sequence information is useful in predicting the biological roles TFs play.

ISSN: 1471-2105