Extended Data Fig. 1: Downstream application adaption method.

Each Downstream application is implemented by segregating the task specific dataset into a training-validation and test (holdout) sets, using the 80:20 split ratio. The training-validation set is further sampled to generate datasets of varying sample size (10% up to 100%). All approaches are trained, separately, on each dataset faction and the resulting models are evaluated and compared on the reserved holdout set.