CPRISP: an interpretable deep learning framework predicting circRNA-RBP interactions and guiding therapeutic discovery in neurodegeneration

Abstract Circular RNAs regulate gene expression through interactions with ribonucleic acid binding proteins, but their binding sites remain difficult to predict across proteins and cellular contexts. Existing methods often focus on transcript features while making limited use of protein-specific information and structural context. Here, we show that an integrated deep learning framework combining circular transcript sequence composition, physicochemical properties, secondary structure, pretrained representations and protein features improves interaction-site prediction across 37 benchmark datasets and generalizes across cell lines; model interpretation recovers conserved single-stranded sequence patterns, while analyses of proteins linked to Alzheimer disease, Parkinson disease and amyotrophic lateral sclerosis identify recurrent pathways involving nucleocytoplasmic transport, lysosomal signaling and guanosine triphosphatase activity and prioritize compounds for experimental follow-up, supporting accurate and interpretable mapping of circular ribonucleic acid protein interactions without implying therapeutic efficacy.