Table 3.
Prediction performance with transfer learning from the primary model to smaller datasets
| Walters et al.’s subset |
Maarastawi et al. |
|||
|---|---|---|---|---|
| Pearson | Bray–Curtis | Pearson | Bray–Curtis | |
| Linear regression | 0.5436 | 0.5596 | 0.1588 | 0.6437 |
| MLP | 0.7114 | 0.4410 | 0.7230 | 0.3347 |
| OTU latent sapce | 0.7544 | 0.4280 | 0.7654 | 0.3233 |
| Combined latent space | 0.7266 | 0.4346 | 0.7060 | 0.3728 |
Note: Walters et al. (2018)’s subset: 100 samples with the same input features as the primary model. Maarastawi et al. (2018): 123 samples with environmental features distinct from those in the primary model. Bold means the best model per column.