Figure 4. Cuticular hydrocarbon (CHC) composition can be used to predict desiccation resistance.

(A) Random Forest regression modeling of CHC abundance was able to explain 85.5% of the variation in time to desiccation with a root mean square error (RMSE) of 4.5. (B) The abundance of four mbCHCs, 2MeC30, 2MeC28, 2MeC32, and 2MeC26, has the highest importance to the desiccation resistance in the random forest regression model, while most of CHCs have less contribution to the accuracy of the model for desiccation resistance.

Figure 4—figure supplement 1. The composition of cuticular hydrocarbons (CHCs) significantly differed across the increasing desiccation resistance.

The non-metric multidimensional scaling (NMDS) plot showing the beta diversity of CHCs differs significantly across the increasing desiccation resistance (permutational multivariate analysis of variance [PERMANOVA]: r² = 0.1, p < 0.001), suggesting CHC composition could affect desiccation resistance in the 50 Drosophila and selected species.

Figure 4—figure supplement 2. Cross-validation of the random forest regression model has a similar performance to the regression model using the full dataset.

A 70:30 training/test split in the dataset (n = 382, n = 164) was used to evaluate the performance of the random forest regression model. The out of bag estimate of the root mean square error (RMSE) in the split dataset is 5.4, which is similar to the RMSE using the full dataset (RNMSE = 4.5).