Inlay

When ranking models by zero-shot prediction performance, we see that most models perform roughly the same and there is much more variation within models among datasets than among models. Also, the phenotype assessed matters for the ranking. 3/