That gap is now being addressed. A research team led by Qiang Kang and Peng Yin of BGI Research, together with collaborators at Dongguan University of Technology and the University of Chinese Academy of Sciences, has unveiled PeTMiner, an end-to-end computational framework designed to mine plant endogenous target mimics by identifying the ternary interactions that link eTMs, microRNAs, and messenger RNAs. The work, published in BMC Genomics, describes a machine learning system that reads raw RNA sequences, learns their deep semantic features, and predicts which three-molecule partnerships are most likely to be biologically real. The study was published on 13 October 2026 as an open-access research article, making the method and its results freely available to the plant science community.

The biological backdrop is the competitive endogenous RNA hypothesis, a framework that has reshaped how biologists think about gene regulation. In this model, transcripts carrying microRNA response elements act as sponges: the more a decoy transcript binds a given microRNA, the fewer microRNA molecules remain free to repress their canonical messenger RNA targets. Endogenous target mimics are a special class of such decoys, characterized by binding sites that sequester microRNAs with high affinity but resist the cleavage that microRNAs normally inflict on their targets. When researchers experimentally disrupt an eTM, the liberated microRNAs surge forward and suppress their mRNA targets, producing measurable shifts in gene expression. Mining eTMs at scale therefore opens a window onto entire regulatory networks, but the search space is enormous and the sequence signals are subtle, which is why previous approaches struggled.

PeTMiner’s architecture tackles the problem in three coordinated stages. First, rather than relying on hand-crafted features such as k-mer frequencies or thermodynamic binding scores, the framework leverages a pre-trained foundation model specialized for plant RNAs. Foundation models of this kind are trained on vast corpora of natural sequences and learn to encode the contextual grammar of RNA molecules, so that each nucleotide is represented in light of its surrounding sequence landscape. By feeding raw eTM, miRNA, and mRNA sequences through this pre-trained encoder, PeTMiner extracts deep semantic representations that capture binding-relevant patterns no human engineer would need to specify by hand. This choice reflects a broader shift in computational biology, where large pre-trained sequence models are increasingly replacing bespoke feature engineering.

The second stage addresses a practical obstacle: RNA sequences, especially those of long non-coding RNAs, can be extremely long, and the token-level representations produced by a foundation model can overwhelm downstream prediction layers. PeTMiner employs a hierarchical attention mechanism to solve this. At the lower level, attention compresses long sequence representations into compact, informative summaries, effectively letting the model decide which stretches of a transcript matter most. At the upper level, attention fuses the representations of the three interacting molecules into a joint embedding, allowing the network to model the cross-molecular compatibility that determines whether a genuine ternary interaction exists. Attention weights also serve an interpretive function, offering clues about which sequence regions drive each prediction, which supports the biological interpretability that the authors emphasize as a key strength of the framework.

The third innovation targets a chronic headache in machine learning applied to genomics: data imbalance. Curated examples of validated eTM-miRNA-mRNA interactions are scarce, and they are unevenly distributed across plant species, with some organisms contributing many known examples and others almost none. A model trained naively on such data tends to optimize for the well-represented species and well-represented interaction classes, quietly sacrificing accuracy everywhere else. PeTMiner counters this with an adaptive weighted loss function that rebalances the optimization process, assigning greater weight to underrepresented examples so that the model does not develop systematic bias toward abundant categories. The result is a classifier whose performance is not an artifact of skewed training data.

How well does it actually work? The authors report extensive experiments across three demanding evaluation regimes: standard test sets drawn from the same distribution as the training data, cross-species test sets in which the model must generalize to plant species it never saw during training, and imbalanced cross-platform test sets that combine species skew with differences in how the data were generated. On all three fronts, PeTMiner achieved superior performance compared with alternative approaches, demonstrating strong generalization ability. This cross-species robustness matters enormously for plant biology, because most crop genomes lack the deep experimental annotation available for model organisms such as Arabidopsis and rice. A tool that transfers knowledge across species can immediately generate hypotheses in orphan genomes where wet-lab validation has barely begun.

Beyond raw predictive accuracy, the framework was designed to feed directly into downstream biology. The eTM candidates that PeTMiner surfaces can be subjected to functional annotation and enrichment analysis, allowing researchers to ask whether the transcripts acting as microRNA decoys cluster around particular biological processes, pathways, or stress-response programs. Because the model’s attention mechanisms provide interpretable signals about the sequence features underlying each prediction, the outputs are not opaque scores but starting points for mechanistic investigation. In principle, a plant biologist could take PeTMiner’s candidate list for a crop species, examine the enriched functions of the predicted decoys, and prioritize experiments on the interactions most likely to influence agronomically important traits.

The implications reach into several arenas of applied plant science. MicroRNA-mediated regulation is deeply entwined with how plants respond to drought, salinity, pathogens, and other stresses, and engineered target mimicry has already been used as a deliberate strategy to tune microRNA activity in transgenic plants. A computational tool that reliably identifies natural eTMs gives researchers a catalog of endogenous tuning knobs, potentially informing breeding programs and synthetic biology designs alike. The method also illustrates how the foundation-model paradigm, borrowed from natural language processing, is maturing within genomics: instead of building a new model from scratch for every prediction task, researchers can now adapt general-purpose RNA encoders to specialized problems like ternary interaction prediction, dramatically lowering the barrier to sophisticated sequence analysis.

Certain caveats remain inherent to the approach. Like any machine learning system, PeTMiner’s predictions are hypotheses that ultimately require experimental confirmation, and the curated data on which such models train reflect the biases of past laboratory work. The authors themselves frame the framework’s output as providing reliable candidates for functional analysis rather than definitive regulatory assignments. Nevertheless, the combination of a plant-specialized foundation model, hierarchical attention for long sequences and cross-molecule fusion, and adaptive loss weighting for imbalanced data represents a substantial technical advance for the competitive endogenous RNA field. As genome assemblies and transcript annotations continue to proliferate across the plant tree of life, tools of this kind will become indispensable for translating raw sequence data into an understanding of the regulatory conversations, and the molecular rivalries, that shape every green organism on Earth.

Subject of Research: Computational prediction of plant endogenous target mimics in competitive endogenous RNA regulation

Article Title: PeTMiner: Mining plant endogenous target mimics via ternary interaction modeling with hierarchical attention and adaptive weighted loss

Article References: PeTMiner: Mining plant endogenous target mimics via ternary interaction modeling with hierarchical attention and adaptive weighted loss. (n.d.). https://doi.org/10.1186/s12864-026-13442-y

Image Credits: AI Generated

DOI: 10.1186/s12864-026-13442-y

Keywords: PeTMiner, endogenous target mimic, competitive endogenous RNA, microRNA, plant genomics, foundation model, attention mechanism, deep learning, gene regulation, BMC Genomics, cross-species prediction, RNA interactions

The central question the researchers posed is deceptively simple: do enhancers share recurring, sequence-based features that distinguish them from the rest of the genome? The team refers to this hypothetical signature as enhancerness. If such a signature exists and can be learned by a machine, it would mean that the grammar of gene regulation is written in the sequence itself, in patterns of bases that recur across species, cell types, and experimental assays. Confirming that would be biologically fundamental, because it would suggest a universal logic underlying how regulatory information is encoded. It would also be enormously practical, since a model that recognizes enhancerness could annotate newly sequenced genomes without waiting for years of costly experiments. The study provides evidence on both fronts, combining computational benchmarks with validation in living flies.

EnhancerDetector is built on a convolutional neural network, the same family of architectures that revolutionized image recognition. Convolutional networks excel at detecting local patterns and assembling them into higher-order structures, which makes them well suited to DNA, where transcription factor binding sites and their combinations form the functional vocabulary of a regulatory element. Trained on human data, the model takes short sequence windows as input and directly outputs a probability that the window is an enhancer. This design choice matters. Many existing enhancer predictors rely on chromatin features, such as histone modifications like H3K4me1 and H3K27ac, which are measured in specific cell types and require complicated post hoc thresholding to convert into enhancer calls. By working directly from sequence, EnhancerDetector sidesteps that dependency and simplifies the discovery workflow, producing scores that can be ranked and filtered without additional processing layers.

The performance results are striking in their breadth. On human, mouse, and fly datasets, EnhancerDetector consistently outperformed existing methods in precision and F1 score, two metrics that matter greatly in genomics because the number of true non-enhancers vastly exceeds the number of true enhancers, making false positives the dominant error. The model also generalized across datasets generated with diverse experimental assays, suggesting it had not merely memorized the artifacts of one laboratory technique. An ensemble strategy, in which multiple models vote together, further improved reliability by reducing false positives. In a field where a single mislabeled enhancer can send a research group down an expensive dead end, that reduction in noise is a meaningful advance. The authors also compared their approach against sequence-based competitors such as gapped k-mer support vector machines, and the convolutional network came out ahead.

Perhaps the most practically important result concerns transfer learning. EnhancerDetector was trained on human data, yet it performed well on mouse and fly, species separated from humans by tens and hundreds of millions of years of evolution respectively. Even more remarkably, the model can be fine-tuned on a new species and retains strong performance when adapted with as few as 20,000 enhancer sequences. For newly sequenced genomes, where experimental data are scarce by definition, this is a game changer. A research consortium that has just assembled the genome of an obscure organism can adapt a pretrained model with a modest dataset and begin prioritizing candidate regulatory elements immediately, rather than starting from scratch. The authors position EnhancerDetector explicitly as a first-stage annotation tool for exactly this scenario.

Accuracy alone, however, was not the goal. The team wanted a model that biologists could interrogate, one whose predictions could be traced back to specific features of the DNA. To achieve this, they applied class activation maps, a visualization technique that highlights which regions of an input sequence contributed most strongly to the model’s decision. These maps allow researchers to see, base by base, where the network is focusing its attention when it declares a stretch of DNA to be an enhancer. Such interpretability transforms the model from a black box into a hypothesis generator: the highlighted motifs can be compared against known transcription factor binding sites, and unexplained patterns can point toward regulatory logic that science has not yet characterized. In an era when machine learning is often criticized for opacity, building interpretability into the architecture from the start is a deliberate and welcome choice.

Computational benchmarks are one thing, but the gold standard in genomics is experimental validation, and here the study delivers. The team tested candidate enhancers predicted by the model in transgenic flies, using reporter constructs in which a candidate DNA sequence is placed next to a green fluorescent protein gene, so that if the sequence acts as an enhancer, the tissue where it is active glows under the microscope. Of six candidates tested, five drove reporter expression, a hit rate that would be respectable even with extensive prior experimental support. Moreover, four of the five exhibited expression patterns consistent with those reported in the earlier literature, providing independent corroboration that the model was not merely finding sequences capable of driving expression in some generic way, but was recovering biologically meaningful regulatory elements with the correct tissue specificities.

Taken together, the computational and experimental results support the paper’s central claim: enhancers possess learnable, recurring sequence features that can be captured in one species and transferred to another. The analyses identified distinct sequence and contextual features associated with enhancer activity, the constellation the authors call enhancerness. This finding resonates with a long-standing debate in regulatory genomics. Enhancers are notoriously diffuse, with binding sites for multiple transcription factors arranged in flexible combinations rather than rigid templates, which has led some researchers to doubt whether any universal sequence signature could exist. The success of a single model trained on human sequence and applied to fly suggests that, beneath the flexibility, there is a shared statistical texture to enhancer DNA, one deep enough for a convolutional network to grasp but subtle enough to have eluded simpler analytical methods.

The implications extend well beyond the three species studied. As sequencing costs continue to fall, thousands of genomes are being assembled for organisms with no experimental annotation at all, from non-model animals to endangered species to agricultural pests. A tool that can flag candidate enhancers from sequence alone, and be fine-tuned with a few tens of thousands of examples, offers a scalable path to functional annotation of this flood of new data. The work was supported by the National Human Genome Research Institute of the National Institutes of Health under award number R21HG011507, and the article is open access, so the tool and its findings are available to any laboratory that wants to build on them. The authors acknowledge assistance with experimental validations from Jack Leatherbarrow and James Weiser, and they declare no competing interests.

There remain limits to what any sequence-only model can do. Enhancer activity depends on cellular context, and a sequence that functions in one tissue may be silent in another, so predictions from DNA alone will always need experimental follow-up. But by demonstrating that a common, learnable signature of enhancerness exists across species and assays, and by packaging that insight into an interpretable, transferable, and experimentally validated framework, the EnhancerDetector team has taken a substantial step toward making the dark matter of the genome a little less dark. For biologists staring at a freshly assembled genome and wondering where the switches are, the answer may now be one pretrained neural network away.

Subject of Research: Cross-species enhancer prediction from DNA sequence using interpretable convolutional neural networks

Article Title: EnhancerDetector: enhancer discovery from human to fly via interpretable deep learning

Article References: Solis, L. M., Sterling-Lentsch, G., Halfon, M. S., & Girgis, H. Z. (2026). EnhancerDetector: enhancer discovery from human to fly via interpretable deep learning. BMC Genomics. https://doi.org/10.1186/s12864-026-13262-0

Image Credits: AI Generated

DOI: 10.1186/s12864-026-13262-0

Keywords: enhancers, deep learning, convolutional neural network, genome annotation, regulatory genomics, cross-species prediction, interpretability, class activation maps, Drosophila, transcription factor binding sites, transfer learning, BMC Genomics