Development and validation of AI/ML derived splice-switching oligonucleotides.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Alternative splicing (AS) is a key mechanism regulating gene expression.
• Splice-switching oligonucleotides (SSOs) are therapeutic agents that modulate AS by binding to pre-mRNA.
• Current methods for identifying functional SSOs can be time-consuming and lack transparency.

Purpose of the Study:

• To demonstrate the utility of artificial intelligence/machine learning (AI/ML) in identifying functional and therapeutic SSOs.
• To enhance transparency in AI/ML-driven drug discovery by predicting targeted splicing factors.
• To prospectively validate an AI/ML-designed SSO for a novel therapeutic target in triple-negative breast cancer (TNBC).

Main Methods:

• Trained XGBoost models using pre-mRNA binding profiles of splicing factors (SFs) and spliceosome assembly data.
• Utilized Shapley and out-of-bag analyses to identify SSO binding sites and predict targeted SFs.
• Applied the AI/ML approach retrospectively to known SSOs and prospectively to NEDD4L exon 13 in TNBC cells.

Main Results:

• AI/ML models successfully identified functional SSO binding sites and predicted targeted SFs, increasing transparency in the discovery process.
• Retrospective analysis validated the AI/ML approach by identifying SFs for previously known functional SSOs.
• A prospectively designed SSO targeting NEDD4L exon 13 reduced TNBC cell proliferation and migration by downregulating the TGFβ pathway.

Conclusions:

• AI/ML provides a powerful and transparent approach for identifying functional SSOs.
• This AI/ML strategy can accelerate drug discovery and yield biological insights.
• Targeting NEDD4L exon 13 with an AI/ML-designed SSO presents a potential therapeutic strategy for TNBC.