Data-driven structure-spectrum mapping of a heterogeneous miRNA ensemble

Short regulatory RNAs populate heterogeneous conformational ensembles in which recurrent local motifs can be difficult to resolve from ensemble-averaged observables. Here, we combine synchrotron circular dichroism (CD), enhanced-sampling molecular dynamics, CD back-calculation, and descriptor-based clustering to map hsa-miR-145 conformations onto their spectral signatures. The hsa-miR-145 duplex provides a reference in which compact A-form-like states produce helical CD profiles, whereas loss of hybridization attenuates and broadens the spectra. In contrast, hsa-miR-145-5p remains highly heterogeneous and lacks persistent self-pairing. Cross-validation across nucleobase-geometry, torsional, and spectral descriptors identifies recurrent compact-bend conformations stabilized by local non-Watson-Crick stacking. These conformers are associated with computed helical-like CD signatures, whereas other recurrent geometries are associated with comparatively flat, low-intensity spectra. The results show that local noncanonical stacking can generate CD-active conformational families in a flexible miRNA strand without canonical pairing.