#14 · ML / AI methods

Computational screening and generative design target RNA splicing and RNAi specificity

The dominant thread is computational prioritization of RNA-directed interventions, with specificity as the central design constraint. eSkip2 ranks exon-skipping ASO target regions across exon-intron contexts using ASO and SNV splicing data. SpeciMiR, trained on 2.2M miRNA-mRNA pairs, generates target-conditioned miRNAs and siRNAs intended to reduce off-target RNAi. A chemoinformatics-guided Tversky screen of more than 8M compounds yields small molecules that bind expanded r(CUG) repeats and rescue MBNL1-dependent splicing in myotonic dystrophy type 1, reported as proof-of-concept. What is new is the application of learned sequence, generative, and similarity-based screening models to the same RNA-level therapeutic problem. The items differ in modality and maturity, so the shared element is methodological rather than a single biological finding.