#16 · Genomics

Learned models predict and design RNA-acting elements with functional readouts

The dominant thread is learned models that take sequence, structure, or experimental data and return a prediction, ranking, or design for an RNA or RNA-acting element. Prediction examples include the ecCLIP atlas and Pythia for RBP-RNA binding and 3' UTR variant effects, LigRNA for RNA-ligand binding sites, SwinTransRNA for lncRNA, circRNA, and miRNA subcellular localization from 6-mer FCGR matrices, and RiboBA for bias-aware ORF identification across ribosome profiling protocols. Design and ranking work targets function: RNAverse generates functional RNAs, and DeSTAR's top-ranked toehold sites yield working sensors about five-fold more often. What is new is the coupling of learned models with direct functional readouts for model-proposed RNAs, evident in DeSTAR and RNAverse but not uniformly across the set. A variational quantum classifier for DNMT3A and DNMT3B flanking methylation motifs applies a related sequence-classification framing to DNA rather than RNA.