Tao Wu, Jingyuan Chen, Wang Lin, Jian Zhan, Mengze Li, Fangzhou Jin, Min Zhang, Kun Kuang, Fei Wu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

RNA language models (LMs) are increasingly applied to RNA structure and function analysis, yet their intrinsic representational capacities remain poorly characterized. Here, we present a standardized zero-shot evaluation of 21 RNA LMs, with representative DNA LMs included as reference controls. Three complementary tasks-attention-based RNA secondary structure prediction, embedding-based RNA classification, and mutational fitness estimation from sequence likelihoods-are evaluated without downstream fine-tuning. Our results reveal substantial variability across models and clear trade-offs between structural, functional, and evolutionary representations. RNA-specific, noncoding RNA-enriched pretraining is crucial for capturing structural information, while evolutionary signals from multiple sequence alignments substantially boost performance. Although model scaling yields gains, architectural and objective choices critically influence performance across task categories. Together, this study provides a foundational benchmark, highlights inherent challenges in learning unified RNA representations, and offers insights for developing next-generation RNA foundation models.

Detection of heart rate and heart rate variability (HRV) is highly important in health monitoring and medical diagnosis, especially under the exercise state. Currently, the main detection techniques include photoplethysmographic volumetric tracing (PPG), electrocardiography (ECG), and ballistocardiogram (BCG). However, ECG and BCG are not suitable for application during exercise, and the accuracy of PPG during exercise is greatly affected by motion artifacts, requiring complex analysis algorithms. Therefore, this paper proposes a novel dynamic-state HRV detection method for a gyrocardiography (GCG)-based chest band with a MEMS inertial measurement unit (IMU). The GCG signals are acquired under three states: resting, jogging, and running. Considering that the actual motion is diverse and may lead to different interference, a novel filter method is proposed after detailed interference measurement and analysis. In addition, the fine peak detection method is proposed to avoid the missed peaks and advance the detection accuracy. The HRV indexes obtained by GCG are compared with those obtained by ECG in the resting state for verification. Meanwhile, the heart rate and HRV indexes obtained by GCG in the jogging and running states are also compared. Experimental results indicate that in the resting state, the accuracy of heart rate measured by GCG is more than 95%, the average accuracy of the HRV indexes is about 92% and the average accuracy in jogging state is 89%. To sum up, the proposed dynamic-state HRV detection method based on a chest band with a MEMS IMU is effective and feasible.

MOTIVATION: The intricate network of RNA-RNA interactions, crucial for orchestrating essential cellular processes like transcriptional and translational regulations, has been unveiling through high-throughput techniques and computational predictions. As experimental determination of RNA-RNA interactions at the base-pair resolution remains challenging, a timely update for assessing complementary computational tools is necessary, particularly given the recent emergence of deep-learning-based methods. RESULTS: Here, we employed base pairs derived from three-dimensional RNA complex structures as a gold standard benchmark to assess the performance of 23 different methods ranging from alignment-based methods, free-energy-based minimization to deep-learning techniques. The result indicates that a deep-learning-based method, SPOT-RNA, can be generalized to make accurate zero-shot predictions of RNA-RNA interactions not only between previously unseen RNA structures but also between RNAs without monomeric structures. The finding underscores the potential of deep learning as a robust tool for advancing our understanding of these complex molecular interactions. AVAILABILITY AND IMPLEMENTATION: All data and codes are available at https://github.com/meilanglang/RNA-RNA-Interaction.

Remarkable effectiveness of the channel or spatial attention mechanisms for producing more discernible feature representation are illustrated in various computer vision tasks. However, modeling the cross-channel relationships with channel dimensionality reduction may bring side effect in extracting deep visual representations. In this paper, a novel efficient multi-scale attention (EMA) module is proposed. Focusing on retaining the information on per channel and decreasing the computational overhead, EMA groups the channel dimensions into multiple sub-features and makes the spatial semantic features well-distributed inside each feature group. Specifically, apart from encoding the global information to re-calibrate the channel-wise weight in each parallel branch, the output features of the two parallel branches are further aggregated by a cross-dimension interaction method. The extensive experiments on common-used benchmarks, such as CIFAR100 for image classification, and object detection on MS COCO and VisDrone2019 datasets, are conducted which indicate that EMA outperforms several recent attention mechanisms significantly without changing networks depth.

Recent advances in deep learning have significantly improved the ability to infer protein sequences directly from protein structures for the fix-backbone design. The methods have evolved from the early use of multi-layer perceptrons to convolutional neural networks, transformers, and graph neural networks (GNN). However, the conventional approach of constructing K-nearest-neighbors (KNN) graph for GNN has limited the utilization of edge information, which plays a critical role in network performance. Here we introduced SPIN-CGNN based on protein contact maps for nearest neighbors. Together with auxiliary edge updates and selective kernels, we found that SPIN-CGNN provided a comparable performance in refolding ability by AlphaFold2 to the current state-of-the-art techniques but a significant improvement over them in term of sequence recovery, perplexity, deviation from amino-acid compositions of native sequences, conservation of hydrophobic positions, and low complexity regions, according to the test by unseen structures, "hallucinated" structures and diffusion models. Results suggest that low complexity regions in the sequences designed by deep learning, for generated structures in particular, remain to be improved, when compared to the native sequences.

Characterizing RNA structures and functions have mostly been focused on 2D, secondary and 3D, tertiary structures. Recent advances in experimental and computational techniques for probing or predicting RNA solvent accessibility make this 1D representation of tertiary structures an increasingly attractive feature to explore. Here, we provide a survey of these recent developments, which indicate the emergence of solvent accessibility as a simple 1D property, adding to secondary and tertiary structures for investigating complex structure-function relations of RNAs.

MOTIVATION: Despite many successes, de novo protein design is not yet a solved problem as its success rate remains low. The low success rate is largely because we do not yet have an accurate energy function for describing the solvent-mediated interaction between amino acid residues in a protein chain. Previous studies showed that an energy function based on series expansions with its parameters optimized for side-chain and loop conformations can lead to one of the most accurate methods for side chain (OSCAR) and loop prediction (LEAP). Following the same strategy, we developed an energy function based on series expansions with the parameters optimized in four separate stages (recovering single-residue types without and with orientation dependence, selecting loop decoys and maintaining the composition of amino acids). We tested the energy function for de novo design by using Monte Carlo simulated annealing. RESULTS: The method for protein design (OSCAR-Design) is found to be as accurate as OSCAR and LEAP for side-chain and loop prediction, respectively. In de novo design, it can recover native residue types ranging from 38% to 43% depending on test sets, conserve hydrophobic/hydrophilic residues at ∼75%, and yield the overall similarity in amino acid compositions at more than 90%. These performance measures are all statistically significantly better than several protein design programs compared. Moreover, the largest hydrophobic patch areas in designed proteins are near or smaller than those in native proteins. Thus, an energy function based on series expansion can be made useful for protein design. AVAILABILITY AND IMPLEMENTATION: The Linux executable version is freely available for academic users at http://zhouyq-lab.szbl.ac.cn/resources/.

MOTIVATION: The accuracy of RNA secondary and tertiary structure prediction can be significantly improved by using structural restraints derived from evolutionary coupling or direct coupling analysis. Currently, these coupling analyses relied on manually curated multiple sequence alignments collected in the Rfam database, which contains 3016 families. By comparison, millions of non-coding RNA sequences are known. Here, we established RNAcmap, a fully automatic pipeline that enables evolutionary coupling analysis for any RNA sequences. The homology search was based on the covariance model built by INFERNAL according to two secondary structure predictors: a folding-based algorithm RNAfold and the latest deep-learning method SPOT-RNA. RESULTS: We showed that the performance of RNAcmap is less dependent on the specific evolutionary coupling tool but is more dependent on the accuracy of secondary structure predictor with the best performance given by RNAcmap (SPOT-RNA). The performance of RNAcmap (SPOT-RNA) is comparable to that based on Rfam-supplied alignment and consistent for those sequences that are not in Rfam collections. Further improvement can be made with a simple meta predictor RNAcmap (SPOT-RNA/RNAfold) depending on which secondary structure predictor can find more homologous sequences. Reliable base-pairing information generated from RNAcmap, for RNAs with high effective homologous sequences, in particular, will be useful for aiding RNA structure prediction. AVAILABILITY AND IMPLEMENTATION: RNAcmap is available as a web server at https://sparks-lab.org/server/rnacmap/ and as a standalone application along with the datasets at https://github.com/sparks-lab-org/RNAcmap_standalone. A platform independent and fully configured docker image of RNAcmap is also provided at https://hub.docker.com/r/jaswindersingh2/rnacmap. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In Minimally Invasive Surgery (MIS), tissue scanning with imaging probes is required for subsurface visualisation to characterise the state of the tissue. However, scanning of large tissue surfaces in the presence of motion is a challenging task for the surgeon. Recently, robot-assisted local tissue scanning has been investigated for motion stabilisation of imaging probes to facilitate the capturing of good quality images and reduce the surgeon's cognitive load. Nonetheless, these approaches require the tissue surface to be static or translating with periodic motion. To eliminate these assumptions, we propose a visual servoing framework for autonomous tissue scanning, able to deal with free-form tissue motion. The 3D structure of the surgical scene is recovered, and a feature-based method is proposed to estimate the motion of the tissue in real-time. The desired scanning trajectory is manually defined on a reference frame and continuously updated using projective geometry to follow the tissue motion and control the movement of the robotic arm. The advantage of the proposed method is that it does not require the learning of the tissue motion prior to scanning and can deal with free-form motion. We deployed this framework on the da Vinci®surgical robot using the da Vinci Research Kit (dVRK) for Ultrasound tissue scanning. Our framework can be easily extended to other probe-based imaging modalities.

Finding information about people using search engines is one of the most common activities on the Web. However, search engines usually return a long list of Web pages, which may be relevant to many namesakes, especially given the explosive growth of Web data. To address the challenge caused by name ambiguity in Web people search, this paper proposes a novel graph-based framework, GRAPE (abbr. a Graph-based fRamework for disAmbiguating People appEarances in Web search). In GRAPE, people tag information (e.g., people name, organization, and email address) surrounding the queried people name is extracted from the search results, a graph-based unsupervised algorithm is then developed to cluster the extracted tags, where a new method, Cohesion, is introduced to measure the importance of a tag for clustering, and each final cluster of tags represents a unique people entity. Experimental results show that our proposed framework outperforms the state-of-the-art Web people name disambiguation approaches.

The ever growing volume of Web data makes it increasingly challenging to accurately find relevant information about a specific person on the Web. To address the challenge caused by name ambiguity in Web people search, this paper explores a novel graph-based framework to both disambiguate and tag people entities in Web search results. Experimental results demonstrate the effectiveness of the proposed framework in tag discovery and name disambiguation.

Adaptive equalization can be used to improve digital data transmission on wireless links with time-varying multipath distortion. It was proved by Zhou, Proakis and Ling (see IEEE Transactions on Communications, vol.38, p.8-24, no.1, 1990) that feedforward schemes are universally capable of approximating any measurable function to any desired degree of accuracy. We propose a new realization of such feedforward scheme to the channel equalization problem, parallel feedforward equalization (PFE). An important feature of the new approach is the decomposition of any equalization into linear and nonlinear components. The new approach chooses F/sub j/(/spl middot/) (j=1, ...) from a family of nonlinear functions to approximate the nonlinear component decomposed from the desired mapping f(/spl middot/). The other new idea proposed in this paper is a measure called nonlinearity distribution which characterizes the nonlinearity in multipath fading channels. The architecture of the new equalization consists of parallel feedforward nonlinear filters, each of them has a specifically tailored nonlinear function F/sub j/(/spl middot/).

We propose a novel class of nonlinear adaptive filters-adaptive neural filters based on a family of nonlinear functions f(/spl middot/) used in neural network, and we introduce several concepts of mapping pairs [X/sub i/,y/sub i/ (i=1,...,N)]. An important idea is to approximate adaptive filters by iteratively searching f/sub i/(/spl middot/): R/sup n//spl rarr/R (i=1,...,K) in N-dimensional topological space, where the f/sub i/(/spl middot/) is a member of the family of nonlinear functions. Another new approach suggested in this paper is to use nonlinear distribution chart, in which the characteristics of nonlinearity in a given approximation (f/sub i/(/spl middot/):R/sup n//spl rarr/R i=1,...,K) can be clearly found, based on the nonlinear distribution charts. There are three steps in the new algorithm: (a) using the least mean-square algorithm to adapt linear approximation with the given approximation pairs [X/sub i/,y/sub i/]: (b) drawing nonlinear distribution chart and using the new algorithm to choose the exact nonlinear function f/sub i/(/spl middot/); (c) adding the hidden nodes with the algorithm. If the established nodes cannot reach the desired precision, until the desired accuracy achieves.>