Sequence similarity filtering reduces data leakage

More stringent filtering of training data reduces cross-similarity with test sets, thereby lowering data leakage and associated model performance inflation.

Part of Chain