HMF-m6A: A Hierarchically Contextual and Multimodally Fused Framework for General RNA m6A Prediction, Validated by HIV-Induced Host Methylation

Accurately predicting RNA m6A methylation and its occurrence probability remains a formidable computational challenge due to the severe interference of local motif contexts, the rigid dependence on prior biological knowledge, and the lack of robust multimodal biophysical constraints. To address these critical bottlenecks, we propose HMF-m6A, a predictive framework independent of prior knowledge that synergistically integrates multimodal features including DNABERT sequence representations, structural topologies, and physicochemical properties through a learnable sparse gating fusion mechanism. By employing an autonomous motif directed routing strategy akin to a mixture of experts, the architecture forces the hierarchical learning of distinct sequence contexts. We further implemented this methodology into a single-step web interface and rigorously evaluated its practical utility in capturing host methylation dynamics during pathogen infections, specifically validated using real-world human immunodeficiency virus (HIV) scenarios. Ablation studies confirmed the functional indispensability of each architectural module. Furthermore, benchmark evaluations demonstrated that HMF-m6A maintains exceptional predictive robustness across diverse motif groups regardless of sample size variations and strictly outperforms five contemporary models across four standard metrics. In the virological application, the framework successfully captured experimentally screened target genes and highly scored methylation samples. The results revealed a strong positive correlation between our predicted continuous probabilities and the actual transcriptomic methylation elevations induced by the viral infection. Ultimately, this accurate and automated architecture provides an extendable computational foundation that precisely reflects dynamic methylation landscapes, offering profound potential for accelerating the discovery of novel epigenetic therapeutic targets in viral pathogenesis.