Antisense oligonucleotides (ASOs) are a powerful class of drugs with the potential to treat a wide range of human diseases. However, the prediction of ASO efficacy remains challenging, as large-scale and costly experimental screens are typically required to identify optimal candidates for a specific therapeutic target. To address this challenge, we compiled ASO Atlas, a database comprising 188,521 RNase H-mediated ASO sequences targeting 334 unique genes with corresponding knockdown efficacy measurements extracted from published patents. Using ASO Atlas, we trained OligoAI, a deep learning model capable of jointly modeling RNA target context, ASO sequence, sugar and backbone chemistries, and dosage to predict in vitro efficacy. We experimentally validated OligoAI by targeting KCNT2 , achieving a 5.72-fold reduction in screening effort compared to random selection. ASO Atlas provides the first systematic resource to rigorously evaluate hypotheses regarding key parameters in ASO design, including sequence composition, chemical modifications, and target region selection. Both ASO Atlas and OligoAI have been made freely accessible through an online web-tool with the aim of facilitating the accelerated optimization of ASO design.