ASOCompass: Context- and Chemistry-Aware Activity Prediction for Transferable Antisense Oligonucleotide Screening

Antisense oligonucleotide (ASO) activity depends on a complex interplay of nucleotide sequence, chemical modifications, target-RNA context, dose, delivery protocol, and cellular environment, yet most computational screening methods only capture a subset of these factors. This limits their ability to predict activity across heterogeneous screening conditions and new biological contexts. The authors introduce ASOCompass, a context- and chemistry-aware framework that integrates contextualized ASO and target-RNA sequence representations with position-specific molecular representations of chemical modifications. It also incorporates dose and delivery information along with prototype-adapted transcriptomic representations of target genes and cell lines, and is jointly trained on auxiliary molecular-property and thermodynamic prediction tasks to encourage chemically informative representations.

Evaluated on ASO Atlas, a large patent-derived dataset of RNase H-mediated gapmer ASOs, ASOCompass achieves an overall Spearman correlation of 0.5970, improving over the strongest ASO-specific baseline by 0.0421, and performs best across all four distribution-shift settings (held-out drug, target gene, cell line, and joint gene-cell line). When adapted to unseen SOD1 and KLKB1 targets, it delivers more accurate candidate ranking across different annotation budgets, reaching correlations of 0.830 and 0.696 with 1,024 target-specific labels. Additional analyses indicate that molecular-property supervision enhances modification-specific ranking, while the auxiliary thermodynamic task helps produce representations aligned with measured inhibition. These results demonstrate that jointly modeling sequence, chemistry, and experimental-biological context can enable more transferable ASO screening.