Table 2.
Summary of 5-fold cross-validation results for ENCODE data
| Model | n | Accur (%) | SD | AUC | # parms | LLexon | LLintron | Ttrain | Tcol |
|---|---|---|---|---|---|---|---|---|---|
| GTR | 0 | 74.8 | 2.6 | .895 | 13 | −11 486 | −85 890 | 3.0 | 1.2×10-5 |
| SH04 | 1 | 77.4 | 2.4 | .918 | 55 | −11 392 | −85 550 | 384 | 2.1×10-4 |
| TRCO+GTR | 1 | 77.8 | 3.9 | .924 | 31 | −11 352 | −85 735 | 10 | 1.3×10-5 |
| ACO+GTR | 1 | 77.3 | 3.9 | .920 | 31 | −11 354 | −85 726 | 382 | 6.2×10-4 |
| RCO+GTR | 1 | 74.9 | 3.2 | .902 | 31 | −11 463 | −85 753 | 26 | 1.9×10-5 |
| MP+GTR | 1 | 74.5 | 2.1 | .859 | 31 | −12 520 | −112 213 | 7 | 1.4×10-5 |
| TRCO+GTR | 2 | 81.2 | 4.2 | .945 | 103 | −11 182 | −85 671 | 180 | 1.7×10-5 |
n: length of context (not including the current column); Accur: mean classification accuracy (percentage correctly classified test cases); SD: standard deviation of accuracy; AUC: area under ROC curve; #parms: number of parameters (including branch lengths); LLexon: mean log-likelihood of training exons; LLintron: mean log-likelihood of training introns; Ttrain: mean training time, in CPU-hours (elapsed time × number of CPUs); Tcol: mean time (seconds) required to evaluate a single column in test alignments.