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

Abstract

(Figure presented.) Machine learning models are trained to identify functional splice-switching oligonucleotides (SSOs) and to predict the splicing factors (SFs) inhibited. SSOs are developed for a novel target in Triple Negative Breast Cancer. Three sources of splicing regulatory information are used to train interpretable XGboost models that predict binding sites for functional SSOs and allow for the identification of SFs regulated by an SSO. The model shows high predictive accuracy in SF-binding perturbation positions critical for AS regulation. A novel alternative splicing target is identified in TNBC, NEDD4L exon 13 (NEDD4Le13). Targeting NEDD4Le13 with the SSO predicted by the model decreases TNBC cell proliferation and migration via downregulation of the TGFβ pathway.