| Single Method | Ensemble Method |
---|
Corpcor | Space | MIND | C&S | C&M | S&M | C&S&M |
---|
EMT | Whole | 35 | 45 | 24 | 45 | 34 | 35 | 41 |
Bootstrap | 32 | 38 | 25 | 40 | 24 | 37 | 40 |
MCC | Whole | 200 | 183 | 210 | 204 | 206 | 201 | 209 |
Bootstrap | 211 | 204 | 207 | 201 | 217 | 220 | 216 |
BR | Whole | 98 | 83 | 95 | 90 | 94 | 97 | 102 |
Bootstrap | 107 | 95 | 99 | 99 | 102 | 100 | 105 |
- The Top 100 correlations for each miRNA were selected from each experiment for performance comparison. To evaluate the effect of three direct correlation inference methods, bootstrapping and Ensemble approach, we performed a comparative study using EMT, MCC and BR datasets. Corpcor (denoted as C) is the partial correlation estimation method, SPACE (denoted as S) is the sparse partial correlation estimation method, and MIND (denoted as M) is the mutual information-based network deconvolution method. ‘Whole’ means that the whole expression profiles were used to infer a direct correlation matrix, and ‘Bootstrap’ means that 100 direct correlation matrices were computed using 100 bootstrapped samples and then aggregated based on an inverse-rank-product method