• RNAinformer uses an Encoder-Decoder to generate RNA sequences given an adjacency matrix (contact map), i.e., like inverse folding
• Use Axial Attention for the adjacency matrix in the encoder and is also capable of conditioning on GC-Content (via linear layers). Use autoregressive decoding for the sequence
• Train on Rfam-derived sequence data, folded by three different algorithms: RNAfold, HotKnots2.0, and RNAformer