Inference of Functional Relations in Predicted Protein Networks with a Machine Learning Approach [Dataset]
Figure_S1: Methods accuracy for the Test Set. The X-axis represents the accumulative number of ‘n’ first predicted interactions, sorted by the corresponding method score, which is different in each case. The Y-axis represents the accuracy, calculated as the ratio of true positives (TP) and total num...
| Autores: | , , , , |
|---|---|
| Tipo de recurso: | conjunto de datos |
| Fecha de publicación: | 2010 |
| País: | España |
| Institución: | Consejo Superior de Investigaciones Científicas (CSIC) |
| Repositorio: | DIGITAL.CSIC. Repositorio Institucional del CSIC |
| OAI Identifier: | oai:dnet:digitalcsic_::6761a440c01a322f799685422070121e |
| Acceso en línea: | http://hdl.handle.net/10261/430413 |
| Access Level: | acceso abierto |
| Palabra clave: | Inference Relations Predicted Networks |
| Sumario: | Figure_S1: Methods accuracy for the Test Set. The X-axis represents the accumulative number of ‘n’ first predicted interactions, sorted by the corresponding method score, which is different in each case. The Y-axis represents the accuracy, calculated as the ratio of true positives (TP) and total number of predictions considered in the test set extracted from our gold standard of functional associations (see Methods). I2H stands for in silico two-hybrid, MT for mirrortree, PP for phylogenetic profiles, GC stands for gene context, GF stands for gene fusion and AODE for Averaged One Dependence Estimators. -- Table S1: Performance of different classifiers for the Test Set. This table shows performance related descriptors for a number of different classifiers. The descriptors included are: Area Under the ROC Curve (AUC), Mathews Correlation Coefficient (MCC, formula shown below), True Positives (TP), True Negatives (TN), False Positives (FP) and False Negatives (FN). The table is divided in three regimes. The first one (yellow background) represents the incremental inclusion of features in AODE classifiers. New features are included from the most to the least discriminative (MCC score for these features) ones: Methods (Gene Fusion, Gene Context, Phylogenetic Profiles, Mirror Tree and in silico two-hybrid), Length (protein sequence lengths) and Nseqs (number of sequences). Rankings in the list of scores for each method are finally included (as they are derived from the corresponding methods) to build the presented APPIA classifier. The second regime (white background) shows the performance for the AODE using all the features and with flags instead of missing values. Finally, the third regime (green background) shows the performance of the other seven different classifying algorithms used in the preliminary test. MCC = (TP×TN−FP×FN)/SQRT((TP + FN)×(TP + FP)×(TN + FP)×(TN + FN)). -- Table S2: Attributes statistics. This table shows some statistical measures of each attribute used in the classification process. The range of the column values represents the minimum and maximum value reached for this attribute in all the examples. In each case, the mean and the standard deviation are calculated without taking into account the instances with an unknown value. Total number of instances: 2,665,180. It should be noted that the high percentage of unknown values is important in many attributes. -- Table S3: List of fully sequenced genomes used. This is the set of 118 prokaryotic genomes to which the computational prediction method has been applied. -- Text S1: AODE detailed description. A detailed description of the AODE algorithm. |
|---|