Experimental techniques for mapping RNA localization, such as subcellular fractionation followed by sequencing or fluorescence-based imaging, are powerful but expensive, slow, and difficult to scale to the thousands of transcripts being discovered by modern transcriptomics projects. This bottleneck has driven a sustained effort to build computational predictors that can read an RNA sequence and infer its destination. Over the past decade, these predictors have evolved from simple models built on hand-crafted features, such as nucleotide composition and sequence motifs, toward deep learning architectures that learn representations directly from raw sequence data. Yet a stubborn problem has persisted: different ways of representing an RNA sequence capture different kinds of information, and no single representation tells the whole story.

A team of researchers at Jingdezhen Ceramic University in Jiangxi, China, has now tackled this integration problem head-on. In a study published in BMC Bioinformatics, Changping Chen, Zi Liu, Wang-Ren Qiu, Liping Zhao, and Xuan Xiao introduce GEPMC-Loc, a dynamic gated ensemble network that fuses pre-trained language models with attention-enhanced multi-scale convolutional feature learning to predict RNA subcellular localization. The work, which appeared on 14 September 2026 as an open-access article, addresses a question that has become central to computational biology in the era of large pre-trained models: when several powerful encoders each see the same biological sequence differently, how do you decide whose opinion to trust?

The architecture of GEPMC-Loc is built around three parallel feature extraction branches, each offering a distinct perspective on the input RNA sequence. The first branch employs ERNIE-RNA, a pre-trained RNA language model that has absorbed statistical regularities from vast collections of RNA sequences. Because of its training regime, ERNIE-RNA produces what the authors describe as structure-aware semantic features, embeddings that implicitly encode information about how nucleotides relate to one another in the folded architecture of the molecule. The second branch uses ProtRNA, an intriguing transfer-learning strategy that borrows representations from a pre-trained protein language model. Although proteins and RNAs are chemically different, protein language models capture evolutionary and physicochemical patterns that, when transferred to RNA sequences, supply complementary information about the biophysical character of the transcript.

The third branch steps away from language models entirely and returns to classical convolutional neural network machinery, but with a twist. An attention-enhanced multi-scale convolutional module scans the sequence with convolutional filters of different sizes, allowing it to detect local sequence patterns at multiple granularities simultaneously. Short filters can pick up compact motifs such as localization signals a few nucleotides long, while wider filters capture longer-range compositional patterns. The attention enhancement lets the module weigh which of these local patterns matter most for a given sequence. Together, the three branches span a spectrum of feature types: semantic and structural knowledge distilled from large-scale pre-training, transferred physicochemical intuition, and explicitly learned local sequence signatures.

The genuinely novel element of GEPMC-Loc, and the source of its name, is the sample-aware dynamic gating mechanism that sits on top of these branches. Rather than fixing the contribution of each branch in advance or averaging their outputs uniformly, the gating network evaluates each input sequence individually and adaptively adjusts how much weight each branch receives in the final integrated representation. For one RNA, the structure-aware embeddings from ERNIE-RNA might dominate; for another, the convolutional branch’s local pattern detection might prove more informative. This per-sample flexibility means the model does not have to commit to a single best encoder, a significant advantage given that lncRNAs, miRNAs, and circRNAs differ dramatically in length, structure, and the nature of their localization signals.

Training such a multi-branch system poses its own challenges, and the authors address them with a multi-objective joint learning strategy. Instead of training only the final fused classifier, GEPMC-Loc optimizes branch-specific learning objectives alongside the ultimate prediction task. Each branch is encouraged to become individually competent at the localization problem, while the gating and fusion layers learn to combine their strengths. This joint optimization prevents a common failure mode of ensemble systems, in which one dominant branch effectively silences the others during training, leaving the architecture with the illusion of diversity but none of its benefits. The result is a network in which all three feature extractors remain active contributors throughout learning.

The researchers evaluated GEPMC-Loc on three benchmark datasets covering the three major RNA classes: long non-coding RNAs, microRNAs, and circular RNAs. Performance was assessed with a battery of standard metrics, including overall Accuracy, Macro-Precision, Macro-Recall, and Macro-F1, the macro-averaged variants being particularly important because they ensure that performance on smaller localization classes is not masked by dominant ones. Across these datasets and metrics, GEPMC-Loc achieved competitive results, demonstrating that the dynamic gating approach generalizes across RNA types rather than being tuned to a single transcript family. The breadth of evaluation matters, because a predictor that works only for one class of RNA offers limited utility as transcriptomics pipelines continue to uncover novel transcripts of every conceivable shape and size.

The significance of this work extends beyond the specific benchmarks. It exemplifies a broader trend in bioinformatics in which pre-trained language models, originally developed in the natural language processing world and later adapted to proteins, are now being repurposed and cross-pollinated across molecular domains. The use of ProtRNA to transfer knowledge from a protein language model to RNA sequences is a particularly striking example of this cross-domain borrowing, betting that the statistical structure learned from one alphabet of life can illuminate another. At the same time, the dynamic gating mechanism offers a template for a recurring design question: as the community accumulates an ever-growing zoo of pre-trained encoders, ensemble frameworks that can weigh their contributions on a per-sample basis may become the standard way to harvest their combined power.

For experimental biologists, accurate computational localization predictions could serve as a first-pass filter, prioritizing which transcripts to study in the lab and generating hypotheses about function for RNAs that have never been characterized. For clinicians and drug developers, the stakes are higher still: many non-coding RNAs are emerging as disease biomarkers and therapeutic targets, and knowing where a candidate RNA operates within the cell shapes both its plausibility as a target and the design of molecules aimed at it. The GEPMC-Loc framework, funded by the National Natural Science Foundation of China and provincial research programs, is available as open-access research, and its authors suggest it provides an effective computational foundation for localization prediction across different RNA types. As the volume of unannotated RNA sequences continues to explode, tools that can read a string of nucleotides and intelligently decide where that molecule belongs in the cell are likely to become indispensable fixtures of the computational biology toolkit.

Subject of Research: Computational prediction of RNA subcellular localization using pre-trained language models and dynamic gated ensemble deep learning

Article Title: GEPMC-Loc: a dynamic gated ensemble network fusing pre-trained language models and multi-scale convolution for RNA subcellular localization

Article References: Chen, C., Liu, Z., Qiu, W.-R., Zhao, L., & Xiao, X. (2026). GEPMC-Loc: a dynamic gated ensemble network fusing pre-trained language models and multi-scale convolution for RNA subcellular localization. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06646-2

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06646-2

Keywords: RNA subcellular localization, pre-trained language models, ERNIE-RNA, ProtRNA, dynamic gating mechanism, multi-scale convolution, deep learning, lncRNA, miRNA, circRNA, bioinformatics, BMC Bioinformatics

The central insight behind the work is that robust expression recognition requires two complementary kinds of visual understanding. The first is local: emotions are encoded in subtle movements of specific facial regions, such as the slight raising of the inner eyebrows that signals worry, or the tightening of the muscles around the eyes that distinguishes a genuine smile from a polite one. The second is global: the spatial arrangement of the entire face, the relationships between regions, and the overall configuration of features all contribute to how an expression is perceived. Most existing architectures emphasize one of these dimensions at the expense of the other. Purely convolutional pipelines excel at extracting local texture patterns but can miss long-range dependencies, while transformer-style models capture global context yet sometimes overlook the fine-grained local cues that carry the emotional signal.

To resolve this tension, the researchers designed a hybrid model that mixes several ResNet-50 branches with VGG-16 backbones. ResNet-50, with its residual connections, allows very deep networks to be trained without the vanishing gradient problems that once stalled deep learning, and it produces rich hierarchical feature maps. VGG-16, an older but still powerful architecture, is prized for its simple stack of small convolution filters that systematically capture texture at fine scale. By fusing representations from both families of networks, the model inherits the strengths of each: the depth-adapted abstractions of ResNet-50 and the fine-textured detail of VGG-16. The authors then attach two purpose-built modules on top of this fused backbone to refine what the networks see and how they interpret it.

The first of these is the Local Feature Enhancement, or LFE, module. Its role is to amplify the subtle variations that occur in the principal facial regions most relevant to emotional expression. In practical terms, the module learns to weight the feature maps so that informative local patterns, such as the curvature of the lips or the degree of eyelid opening, are emphasized while less relevant background or redundant information is suppressed. This kind of learned emphasis matters enormously in expression recognition, where the difference between two emotions can hinge on a few pixels. Without such focusing, a network may spread its capacity evenly across the image and fail to register the discriminative details that separate sadness from neutrality in a low-resolution photograph.

The second component, the Global Information Association, or GIA, module, works at the opposite end of the spatial scale. It captures the global structure of the face, modeling how distant regions of the image relate to one another. An expression is not merely a sum of parts; a raised mouth corner reads as joy only in combination with relaxed eyes, whereas the same mouth movement alongside furrowed brows may indicate a different state entirely. By associating information across the whole face, the GIA module gives the network a holistic understanding of the expression, allowing it to integrate the locally enhanced features into a coherent global interpretation.

Between these two modules, the architecture deploys a multi-scale convolutional design together with a self-attention-based approach known as ACMix. Multi-scale convolution means that the network examines the face simultaneously at several receptive field sizes, from small windows that catch fine wrinkles to larger windows that span whole facial regions. This mirrors the way human vision processes faces at multiple resolutions, and it helps the model remain accurate when faces appear at different sizes or distances. The self-attention mechanism, meanwhile, allows every part of the feature representation to dynamically attend to every other part, computing relevance weights on the fly. ACMix blends convolutional inductive biases with attention-style global reasoning, which improves the model’s stability under the challenging conditions, such as unusual illumination or partial occlusion, that routinely break less flexible architectures.

Evaluation was carried out on three of the most widely used benchmarks in the field. RAF-DB contains thousands of real-world, in-the-wild faces annotated with expressions, capturing the messy variety of unconstrained photography. FER2013, the dataset used for training and evaluation, is freely available through Kaggle and comprises more than 35,000 grayscale images cataloged across seven emotion categories: anger, disgust, fear, happiness, sadness, surprise and neutral expressions, including multiple views of the same subjects, making it the largest such resource of its kind. CK+ offers carefully controlled laboratory sequences with well-defined expressions, providing a complementary test of discriminative power. Achieving 95.87 percent accuracy across these tests places the hybrid model among the strongest performers reported in the literature, and the authors emphasize that its advantage grows when conditions become difficult, such as when glasses, scarves or hands obscure parts of the face, or when the head is rotated away from the camera.

Particularly notable is the model’s behavior in cross-dataset testing, one of the most demanding evaluations in expression recognition. A system trained on one dataset and tested on another must cope with differences in lighting, demographics, image quality and labeling conventions, and accuracy typically drops sharply in such settings. The Moroccan team reports that their model shows superior robustness under exactly these conditions, suggesting that the combination of local enhancement, global association and attention produces features that generalize rather than merely memorize dataset-specific quirks. This generalization is precisely the property that separates laboratory demonstrations from technology that can function reliably outside controlled environments, and it is a key reason the authors argue their approach is suited to real-world applications.

The potential applications span several domains where reading a face accurately is not a luxury but a safety requirement. In healthcare, precise identification of emotions can support monitoring of patients who cannot communicate verbally, alerting clinicians to pain or distress. In driver monitoring systems, a camera that reliably detects fatigue, distraction or confusion even when the driver’s face is partly occluded or turned could help prevent accidents. The authors point to these use cases, where missing a critical emotional state carries real consequences, as the natural fit for their architecture. The work was conducted with institutional resources at Ibn Tofail University and Sidi Mohamed Ben Abdellah University in Morocco, without external funding, and the datasets underpinning the study are openly available under appropriate licenses, which should make it straightforward for other research groups to reproduce and extend the results.

As facial expression recognition moves from the research bench into cars, clinics and everyday devices, studies like this one illustrate the recipe that is emerging as the field’s consensus: hybrid backbones that unite complementary convolutional designs, explicit modules for local detail and global context, and attention mechanisms that let the network decide what matters in each image. By combining ResNet-50 and VGG-16 with the LFE, GIA and ACMix components, the Moroccan team has demonstrated that this layered strategy can push accuracy toward 96 percent while keeping the model dependable when the real world, with its shadows, angles and obstructions, refuses to cooperate. The full details of the architecture and experiments are available in the article published in Multimedia Tools and Applications, volume 85, article number 775.

Subject of Research: Hybrid deep learning with attention mechanisms for robust facial expression recognition

Article Title: Hybrid convolutional neural network with attention mechanism to provide robust face expression recognition

Article References: Ouza, A., Ghmary, M. E., & Choukri, A. (2026). Hybrid convolutional neural network with attention mechanism to provide robust face expression recognition. Multimedia Tools and Applications, 85(10), Article 775. https://doi.org/10.1007/s11042-026-21933-z

Image Credits: AI Generated

DOI: 10.1007/s11042-026-21933-z

Keywords: facial expression recognition, deep learning, convolutional neural networks, attention mechanism, ResNet-50, VGG-16, multi-scale convolution, occlusion, pose variation, RAF-DB, FER2013, computer vision