Table 2. Comparison of models for predicting neighbour preferences.
|
Model* |
||||||
|---|---|---|---|---|---|---|
| Triplet | 1st 5′ |
Multiplicative |
||||
| 1st 5′and 1st 3′ | 1st+2nd 5′ and 1st+2nd 3′ | 1st−3rd 5′ and 1st−3rd 3′ | 1st−4th 5′ and 1st−4th 3′ | |||
| hADAR1 | 59.2% | 52.8% | 59.0% | 69.5% | 73.0% | 77.1% |
| hADAR1-D | 66.5% | 54.2% | 66.8% | 78.6% | 83.6% | 86.4% |
| hADAR2 | 45.3% | 35.0% | 44.8% | 47.5% | 52.1% | 57.0% |
| hADAR2-D | 45.4% | 37.7% | 45.6% | 48.2% | 57.7% | 60.4% |
| Model # |
1 |
2 |
3 |
4 |
5 |
6 |
*Percentages are adjusted R2 values. The triplet model (leftmost column of numbers) estimates the % editing of the target adenosine based on the immediate neighbouring 5′ and 3′ bases. This model includes 16 different coefficients to allow the effect of the neighbouring 5′ base to depend on the identity of the neighbouring 3′ base, and conversely, allows the effect of the neighbouring 3′ base to depend on the identity of the neighbouring 5′ base. The remaining models estimate the % editing of the target adenosine based on the identities of 1, 2, 3 or 4 bases on the 5′ and 3′ sides. In contrast to the triplet model, each of the remaining models achieves increased parsimony by invoking the simplifying assumption that the effect of a base at a particular position is not altered by the identities of the bases at other positions.