Table 2. Comparison of 10 programs for RNA inverse folding, benchmarked on 63 target structures, as explained in the text.
| Method | ERD | FRNA | Incarnation | Info-RNA | MODENA | Nupack | RNA-SSD | RNAfbinv | RNAiFold2 | RNAinverse |
|---|---|---|---|---|---|---|---|---|---|---|
| Output (%) | 100% | 30% | 60% | 95% | 60% | 57% | 90% | 13% | 65% | 65% |
| Target (%) | 85% | 38% | 0% | 57% | 45% | 70% | 82% | 0% | 100% | 18% |
| Avg str len | 397 | 122 | 352 | 393 | 234 | 256 | 400 | 74 | 363 | 208 |
| Avg output | 117 | 325 | 41 535 | 195 | 50 | 22 | 1 | 2 | 55,476 | 935 |
| P(S) | 3.32% | 1.70% | 0.06% | 3.17% | 11.30% | 30.01% | 2.24% | 0.36% | 23.21% | 0.78% |
| Native cont. (%) | 85 ± 9 | 61 ± 15 | 63 ± 13 | 76 ± 12 | 89 ± 9 | 98 ± 1 | 85 | 32 ± 6 | 93 ± 2 | 57 ± 12 |
| Avg E | −0.41 | −0.24 | −0.46 | −0.63 | −0.46 | −0.44 | −0.30 | −0.14 | −0.56 | −0.23 |
| Pos entropy | 0.33 | 0.71 | 0.41 | 0.44 | 0.15 | 0.07 | 0.36 | 0.88 | 0.12 | 0.80 |
| MH diversity | 0.16 | 0.35 | 0.21 | 0.22 | 0.07 | 0.03 | 0.18 | 0.45 | 0.06 | 0.38 |
| Vienna diversity | 0.11 | 0.23 | 0.15 | 0.16 | 0.05 | 0.02 | 0.11 | 0.30 | 0.05 | 0.26 |
| Exp bp dist | 0.09 | 0.21 | 0.27 | 0.16 | 0.06 | 0.01 | 0.08 | 0.38 | 0.03 | 0.24 |
| Ens def | 0.14 | 0.32 | 0.39 | 0.22 | 0.08 | 0.02 | 0.14 | 0.56 | 0.04 | 0.37 |
| Exp num bp | 0.28 | 0.29 | 0.34 | 0.30 | 0.26 | 0.29 | 0.28 | 0.28 | 0.27 | 0.28 |
| GC-content (%) | 55% | 49% | 71% | 72% | 50% | 57% | 36% | 51% | 57% | 49% |
Averages are given, rounded either to two decimals or to the nearest integer as appropriate. Complete data, with averages and standard deviations, can be found on the web server RNAiFold 2.0. FRNA stands for FRNAnkenstein. Row labels are as follows, whereby measures appearing after the double line have been normalized by dividing by sequence length—for instance, Avg E denotes the normalized average free energy of the returned sequences, computed as the average, taken over all 63 individual target structures S0, of average normalized free energies E(a, S0)/|a|, taken over all sequences a returned for target structure S0, where E(a, S0) denotes the free energy of sequence a with respect to the structure S0. The other normalized measures are defined in an analogous manner. (Unnormalized measures) Output (%): fraction of the 63 target structures for which some output was produced. Target (%): average fraction of output sequences whose MFE structure is the target. Avg str len: average target structure length, taken over those target structures for which at least one output sequence was returned. Avg output: total number of sequences returned for all 63 targets, divided by the number of targets for which at least one sequence was returned. P(S): average probability of target structure, defined as the average, taken over all 63 target structures S0, of the average Boltzmann probability P(s, S0) = (exp ( −E(s, S0)/RT)/Z, taken over all sequences s returned for target structure S0. (Normalized measures) Avg E: normalized average free energy with respect to target (previously defined). The remain measures are length-normalized versions of positional entropy, Morgan-Higgs diversity, Vienna diversity, expected base pair distance from target structure, ensemble defect with respect to target structure, expected number of base pairs, proportion of native contacts and GC-content. Measures are defined in the text.