Untranslated Region-FNex: An Integrative Multimodal Framework for Functional Untranslated Region Variant Prioritization

Abstract

Genetic variants in untranslated regions (UTRs) can alter post-transcriptional regulation, yet functional interpretation remains difficult because regulatory mechanisms differ between 5′UTR and 3′UTR contexts and are not fully captured by generic noncoding-variant scores. We developed UTR-FNex, an interpretable machine-learning framework for prioritizing candidate functional variants in both UTR regions. We first constructed MetaUTR, a harmonized resource of approximately 200,000 UTR variant loci, and derived model-development sets containing 14,627 3′UTR and 3533 5′UTR variants after confidence-based filtering. UTR-FNex integrates region-specific sequence, RNA-structure, regulatory-motif, miRNA, conservation, and DNABERT-derived allele representations, with a neural feature enhancement (NFE) module providing additional interaction summaries. Across 10 repeated random hold-outs, mean AUROCs were 0.876 for 3′UTR and 0.974 for 5′UTR. Under gene-grouped evaluation, AUROCs decreased to 0.769 and 0.672, respectively, indicating substantial context dependence, especially for 5′UTR. In an external prostate-cancer 3′UTR cohort, UTR-FNex achieved an AUROC of 0.720 (95% CI 0.690–0.750), compared with 0.696 (95% CI 0.660–0.733) for FunUV; however, FunUV achieved higher AUPRC and stronger top-k enrichment. A TP53 rs78378222 case study connected model sensitivity to a known polyadenylation-signal context, whereas RNAfold-derived differences were window-dependent and were treated as exploratory. These results support UTR-FNex as a multimodal prioritization framework while defining important limits in cross-gene and cross-source generalization. Source code is available at https://github.com/liyaning233/UTR-FNex, and MetaUTR is deposited at 10.5281/zenodo.19719670.