Fig. 4: RNA family classification performance through dimensionality reduction analysis.
From: ERNIE-RNA: an RNA language model with structure-enhanced representations

t-SNE visualization comparing clustering results using different feature representations: one-hot encoding (top left), RNA-FM CLS token embeddings (top right), ERNIE-RNA CLS token embeddings (bottom left), and ERNIE-RNA attention maps (bottom right). Each color represents a distinct RNA family category. The Rand Index and Fowlkes-Mallows scores, displayed in the top-left corner of each panel, quantitatively measure the clustering quality.