| | Nb annotated probes | Nb unique genes | R2 | RMSE |
|---|
RFI-BV | 384 | 222 | 0.63 | 42.9 |
280 | 161 | 0.64 | 39.6 |
50 | 27 | 0.65 | 39.3 |
FCR | 421 | 267 | 0.61 | 0.23 |
88 | 52 | 0.70 | 0.22 |
50 | 33 | 0.67 | 0.22 |
FCRe | 318 | 218 | 0.49 | 2.2 |
50 | 29 | 0.52 | 2.0 |
| | 7 | 6 | 0.52 | 2.0 |
- Machine learning procedure (gradient tree net boosting) was applied on microarrays dataset (20,405 expressed annotated probes) generated from the longissimus muscle of 71 growing pigs to identify models able to predict residual feed intake (RFI), feed conversion ratio (FCR) and net energy-based feed conversion ratio (FCRe). A randomly selected bootstrap pig sample (n = 50) was used for learning, whereas the remaining pigs (n = 21) was used for validation test. The first rounds led to model stabilization with 384 molecular probes as very important variables (VIP) for RFI-BV prediction, 421 probes for FCR prediction and 318 probes for FCRe prediction, respectively, out of the 20,405 expressed annotated probes. The second entry was an iterative step of the former procedure but considering the VIP that were identified in the first step as the new inputs. This increased the accuracy of the prediction evaluated by the root mean square error (RMSE) and the coefficient of determination (R2). The last entry was another iterative step using the VIP identified at the second step as the new inputs, which led to identify the smallest number of VIP able to predict the target trait with a good accuracy. The numbers (Nb) of annotated probes and their corresponding unique genes identified as VIP for the three feed efficiency traits were indicated. Lists of the VIP (probes and their corresponding gene name when applicable) are provided in Additional files 1, 2 and 3
- TreeNet boosting procedure was applied to 20,405 annotated probes expressed in the longissimus muscle of 71 pigs to release very important predictors (VIP) that can be used to predict residual feed intake (RFI) values. A total of 384 molecular probes were identified. Iterative steps led to reduce the set to 50 molecular probes corresponding to 30 unique encoded genes. These genes were listed by the order of importance (score) in prediction. Expression levels of genes indicated in bold face were further measured by qPCR
