Fig 1. Systematic increase of training data size improves prediction quality.

The experimental data on the relative fitness of E. coli with mutations in β-lactamase was modeled. The fraction of the complete data that was used for training and validation was systematically reduced in four steps from 85% to 15% to see how the quality of computational predictions of fitness changes. It can be seen that the quality of predictions when trained with 50% is comparable with the one trained at 85% data. The prediction quality is tabulated in Table 1 in S1 File. Results from predictions of other proteins are in Figs 3-5 in S2 File.