ASO-RASAR: A Read-Across Framework for Predicting Antisense Oligonucleotide Gapmer Activity Across Target Genes

Area of Science:

• Drug Discovery
• Bioinformatics
• Oligonucleotide Therapeutics

Background:

• Predictive modeling for antisense oligonucleotide (ASO) drug discovery is crucial but faces challenges in generalizing across diverse target genes.
• Existing sequence-activity relationship (QSAR) models often fail to predict ASO efficacy for novel targets due to target-specific determinants.

Purpose of the Study:

• To develop a novel predictive model, ASO-RASAR, that transfers sequence-activity relationship information from data-rich to data-poor target genes.
• To address limitations in ASO candidate prioritization for targets with scarce experimental data.

Main Methods:

• Curated a dataset of 59,273 gapmer ASOs across 30 human genes.
• Benchmarked gene-specific and cross-target prediction models using sequence and target-context descriptors.
• Developed and validated the ASO-RASAR read-across model for predicting ASO activity in low-data scenarios.

Main Results:

• Gene-specific models showed strong performance, with genomic context and sequence motifs being key features.
• Cross-target models demonstrated poor generalization to new genes, highlighting target-specific activity.
• ASO-RASAR improved prediction performance by up to 22.5% in simulated low-data settings.
• Experimental validation showed ASO-RASAR scores strongly correlated with knockdown efficiency (Pearson's r = 0.83).

Conclusions:

• ASO activity is primarily governed by target-specific factors, limiting generalizability of standard models.
• ASO-RASAR provides a practical solution for prioritizing ASO candidates against novel targets with limited data.
• This approach enhances the efficiency of ASO drug discovery by leveraging existing data effectively.