Discovering molecular features of intrinsically disordered regions by using evolution for contrastive learning
Fig 2
UMAP scatterplot of reverse homology features for our yeast model.
Reverse homology features are extracted using the final convolutional layer of the target encoder: max-pooled features are shown in red, while average-pooled features are shown in blue. We show the sequence logo corresponding to select features, named using the index at which they occur in our architecture (see Methods for how these are generated). Amino acids are colored according to their property, as shown by the legend at the bottom. All sequence logos range from 0 to 4.0 bits on the y-axis.