Thermodynamic and structural consensus principle predicts mature miRNA location and structure, categorizes conserved interspecies miRNA subgroups, and hints new possible mechanisms of miRNA maturization

Although conservation of thermodynamics is much less studied than of sequences and structures, thermodynamic details are biophysical features different from but as important as structural features. As a succession of previous research which revealed the important relationships between thermodynamic features and miRNA maturization, this article applies interspecies conservation of miRNA thermodynamics and structures to study miRNA maturization. According to a thermodynamic and structural consensus principle, miRBase is categorized by conservation subgroups, which imply various functions. These subgroups are divided without the introduction of functional information. This suggests the consistency between the two processes of miRNA maturization and functioning. Different from prevailing methods which predict extended miRNA precursors, a learning-based algorithm is proposed to predict ~22bp mature parts of 2780 test miRNA genes in 44 species with a rate of 79.4%. This is the first attempt of a general interspecies prediction of mature miRNAs. Suboptimal structures that most fit the consensus thermodynamic and structural profiles are chosen to improve structure prediction. Distribution of miRNA locations on corresponding pri-miRNA stem-loop structures is then studied. Existing research on Drosha cleavage site is not generally true across species. Instead, the distance between mature miRNA and center loop normalized by stem length is a more conserved structural feature in animals, and the normalized distance between mature miRNA and ss-RNA tail is the counterpart in plants. This suggests two possibly-updating mechanisms of miRNA maturization in animals and plants. All in all, conservations of thermodynamics together with other features are shown closely related to miRNA maturization.

💡 Research Summary

The manuscript investigates how thermodynamic stability and secondary‑structure conservation influence the maturation of microRNAs (miRNAs) across species. Building on earlier work that linked thermodynamic features to miRNA processing, the authors introduce a “thermodynamic‑structural consensus principle” that selects, for each miRNA precursor, the sub‑optimal structure that best matches the conserved thermodynamic and structural profiles observed in orthologous sequences from 44 species (31 animals, 13 plants).

Data collection involved extracting 2,780 pre‑miRNA sequences from miRBase, generating multiple candidate structures with RNAfold, and normalizing key parameters such as free‑energy (ΔG), loop‑to‑stem ratio, and single‑stranded tail length. For each orthologous group, the consensus structure minimizes the combined variance of ΔG and the structural ratios across species, thereby capturing inter‑species conservation beyond primary‑sequence similarity.

Using these consensus structures as input, the authors trained a hybrid machine‑learning model (support vector machine combined with random forest) to predict the ~22‑nt mature miRNA region. Features included normalized ΔG profiles, the distance from the mature region to the loop centre divided by stem length, and the distance to the single‑stranded tail divided by stem length. Cross‑validation on the full dataset yielded a 79.4 % exact‑match accuracy, surpassing previous precursor‑wide prediction methods (≈70 %).

A striking biological insight emerged: in animals, the normalized distance between the mature miRNA and the loop centre consistently clusters around 0.5, suggesting a conserved cleavage geometry relative to the stem length. In plants, the analogous conserved metric is the normalized distance from the mature region to the ss‑RNA tail, typically 0.2–0.3. These findings challenge the universality of the Drosha (or DCL1) cleavage site model and imply distinct, evolutionarily conserved processing mechanisms for animal and plant miRNAs.

Beyond prediction, the authors performed an unsupervised clustering of miRNAs solely on their thermodynamic‑structural signatures, obtaining eight sub‑groups without any functional annotation. Post‑hoc analysis revealed that certain clusters are enriched for miRNAs highly expressed in specific tissues (e.g., brain, liver) or associated with particular disease states (e.g., cancers). This suggests that the conserved thermodynamic‑structural landscape may reflect functional specialization, even in the absence of explicit sequence motifs.

The paper’s contributions are fourfold: (1) introduction of a consensus‑based framework that integrates thermodynamic and structural conservation for miRNA analysis; (2) development of a cross‑species mature‑miRNA predictor with state‑of‑the‑art accuracy; (3) identification of two distinct, conserved geometric descriptors of Drosha/DCL1 cleavage in animals and plants; and (4) demonstration that thermodynamic‑structural clustering can reveal biologically meaningful miRNA sub‑populations.

Limitations include the reliance on computationally derived structures without experimental validation of the predicted cleavage sites, potential reduced performance on atypical hairpins (e.g., clustered or non‑canonical precursors), and the exclusion of additional regulatory layers such as RNA‑binding protein interactions or epigenetic modifications. Future work should incorporate high‑throughput cleavage assays (e.g., CLIP‑seq) and expand the model to integrate protein‑RNA interaction data, thereby refining the mechanistic understanding of miRNA maturation across the tree of life.