Biology subjects

Zheng, J.

Publications and source records attributed to Zheng, J..

At least 91 records · Page 5Linked to original sources

bioRxiv2025

Mitochondrial heterogeneity disrupts osteoclast differentiation and bone resorption by impairing respiratory complex I

Mitochondrial heteroplasmy, the co-existence of different mitochondrial genomes within a cell, is linked to aging and disease. Patients with heteroplasmy due to mitochondrial mutations experience multiple organ complications, particularly poor bone health and bone structure defects. However, the mechanisms involved are generally unknown, due largely to the difficulty of manipulating mtDNA in vivo. To overcome this, we leveraged a heteroplasmic mouse model and discovered that mitochondrial heteroplasmy affects a fundamental developmental process. Specifically, the differentiation of osteoclasts, which resorb bone tissue and maintain bone homeostasis. Mechanistically, there was a reduced localization of specifically respiratory complex I subunits in mitochondria in heteroplasmic mice, disrupting ATP production and osteoclast differentiation. In addition, autophagic flux is exhausted, and the autophagy inducer spermidine restores mitochondrial health and rescues osteoclast activity, both in mice and in cells from patients with primary mitochondrial disease. Together, we identify the mechanisms by which mitochondrial heteroplasmy impacts osteoclastogenesis and discover spermidine as a modulator of this process, which presents a potential treatment for human heteroplasmic conditions such as mitochondrial diseases, which are largely untreatable. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=152 SRC="FIGDIR/small/651799v4_ufig1.gif" ALT="Figure 1"> View larger version (94K): org.highwire.dtl.DTLVardef@131d3f6org.highwire.dtl.DTLVardef@52bf2aorg.highwire.dtl.DTLVardef@848bb3org.highwire.dtl.DTLVardef@1efa572_HPS_FORMAT_FIGEXP M_FIG C_FIG

cell biology↗

bioRxiv2025

A large global soil carbon sink informed by repeated soil samplings

Partitioning the terrestrial carbon sink between vegetation and soil is crucial for predicting future climate change, but the role of soils remains poorly quantified. Here, we compiled 3,099 soil organic carbon time series spanning five decades. We found a global soil organic carbon sink of 1.83 {+/-} 0.9 (mean {+/-} SE) petagrams per year from 1992 to 2020, driven by extratropical young forests, boreal old forests, and grasslands, while trends in tropical ecosystems remain uncertain. Our findings suggest the net land sink resides almost exclusively belowground as soil carbon, emphasizing the global opportunity of soil conservation and restoration for climate mitigation.

ecology↗

bioRxiv2025

Unique Gluing Effect of ASXL1 K351 Monoubiquitination Stimulates PR-DUB-Mediated Nucleosomal H2AK119Ub Deubiquitination

Protein ubiquitination plays a critical role in regulating fundamental biological processes through both proteolytic and nonproteolytic mechanisms. While classically known for its role in targeting proteins for degradation, ubiquitination can also modulate enzymatic activity. However, all of the known mechanisms of such modulation involve spatially constrained mechanisms that take place near the enzyme-substrate interfaces. Here, we report an unprecedented regulatory paradigm mediated by ubiquitination that is located distal to the enzyme-substrate interface but can still activate the polycomb repressive deubiquitinase (PR-DUB) complex. Using cryo-electron microscopy, molecular dynamics (MD) simulations, hydrogen-deuterium exchange mass spectrometry (HDX-MS), and kinetics assays, we determined that the ASXL1 K351 monoubiquitination promotes PR-DUB-mediated nucleosomal H2AK119Ub deubiquitination by stabilizing the catalytic pocket of PR-DUB, causing significant increase in maximal catalytic velocity (Vmax) but almost no change in the substrate binding affinity (Km). Structural analysis revealed that the ubiquitin moiety at ASXL1 K351 bridges the BAP1 and ASXL1 subunits, acting as an intramolecular "glue" that constrains the conformational dynamics of these subunits without altering the nucleosome binding interface. Further MD and HDX-MS results demonstrated the monoubiquitination of ASXL1 K351 enables PR-DUB to effectively maintain a spatial arrangement of the catalytic conformation that closely aligns with substrate cleavage. This study expands our understanding of the mechanisms of ubiquitin function to include intramolecular fastening with a molecular gluing effect and provides a deeper mechanistic understanding of PR-DUB activation in nucleosomal H2AK119Ub removal.

biochemistry↗

bioRxiv2025

MiT4SL: multi-omics triplet representation learning for cancer cell line-adapted prediction of synthetic lethality

Synthetic lethality (SL) offers a promising approach for targeted cancer therapies. Current SL prediction models heavily rely on extensive labeled data for specific cell lines to accurately identify SL pairs. However, a major limitation is the scarcity of SL labels across most cell lines, which makes it challenging to predict SL pairs for target cell lines with limited or even no available labels in real-world scenarios. Furthermore, gene interactions could be opposite between training and test cell lines, i.e. SL vs. non-SL, which further aggravates the challenge of generalization among cell lines. A promising strategy is to transfer knowledge learned from cell lines with relatively abundant SL labels to those with limited SL labels for the discovery of novel SL pairs, i.e., cell line-adapted SL prediction. Here, we propose MiT4SL, a multi-omics triplet representation learning model for cell line-adapted SL prediction. The core idea of MiT4SL is to model cell lineage information as embeddings, which are generated by combining a protein-protein interaction network representation tailored to each cell line with the corresponding protein sequence embeddings. We then combine these cell line embeddings with gene pair representations derived from a biomedical knowledge graph and protein sequences. This triplet representation learning strategy enables MiT4SL to capture both shared biological mechanisms across cell lines and those unique to each cell line, effectively mitigating distribution shift and improving generalization to target cell lines. Additionally, explicit cell line embeddings provide the necessary signals for MiT4SL to effectively differentiate between cell line contexts, enabling it to adjust predictions and mitigate possible label conflicts for the same gene pair across different cell lines. Experimental results across various cell line-adapted scenarios show that MiT4SL outperforms six state-of-the-art models. To the best of our knowledge, MiT4SL is the first deep learning model designed specifically for cancer cell line-adapted SL prediction. AvailabilityThe code of our work is available at https://github.com/JieZheng-ShanghaiTech/MiT4SL.

bioinformatics↗

bioRxiv2025

Ontology-based Protein-Protein Interaction Explanation Using Large Language Models

Protein-protein interactions (PPIs) play a crucial role in various biological processes, and understanding these interactions is essential for advancing biomedical research. Automated extraction and analysis of PPI information from the rapidly growing scientific literature remains an important challenge. We present a novel ontology-based approach to analyze protein-protein interactions using Large Language Models (LLMs). We applied different learning strategies, namely in-context learning and parameter-efficient instruction fine-tuning for the Llama-2 chat models, to identify keywords in the text that indicate an interaction between a pair of proteins. Our results show that parameter-efficient fine-tuning leads to a performance gain even when the domain is new. The smaller fine-tuned models outperformed the zero-shot performance of much larger models. The keywords identified by the Llama-2 models were mapped to the ontology terms in the Interaction Network Ontology (INO). Our study suggests that a pipeline of an LLM and an ontology is an effective strategy for explaining relations between biomedical entities. This work demonstrates the potential of leveraging ontologies and advanced language models to advance automated PPI analysis from the scientific literature.

bioinformatics↗

bioRxiv2025

Allo-PED: Leveraging protein language models and structure features for allosteric site prediction

Allosteric regulation plays a pivotal role in modulating protein function and allosteric sites represent a promising target for drug discovery. However, identifying allosteric sites remains challenging due to their structural and evolutionary diversity. Here, we present AlloPED, a novel framework that combines protein language models and machine learning to predict allosteric sites with high accuracy. AlloPED consists of two modules: AlloPED-pocket, an ensemble model leveraging physicochemical features to predict allosteric pockets; and AlloPED-site, a dilated convolutional neural network (DCNN) augmented with a comprehensive attention mechanism for residue-level prediction. AlloPED-pocket achieves state-of-the-art performance on benchmark datasets, yielding an MCC of 0.544 and an AUC of 0.920, outperforming existing methods such as AllositePro and PARS. AlloPED-site further refines predictions using high-dimensional sequence embeddings from the ProtT5 protein language model, achieving a precision of 0.601, a recall of 0.422, and a specificity of 0.661. These results highlight the effectiveness of integrating ensemble learning and deep learning for allosteric site prediction. AlloPED also identifies critical determinants of allosteric sites, including residue clustering coefficients, van der Waals volume, and hydrophobic microenvironments. In summary, this framework provides a robust tool for advancing our understanding of allosteric regulation and facilitating structure-based drug design.

bioinformatics↗

bioRxiv2025

Effect of Shufeng Xingbi Therapy on Th1/Th2 immune balance and intestinal flora in rats with allergic rhinitis

ObjectiveTo study the effect of Shufeng Xingbi Therapy (SFXBT) on Th1/Th2 immune balance and intestinal flora in allergic rhinitis (AR) rats. Methods32 SD rats were randomly divided into control group, OVA group, antibiotic + SFXBT group and acetic acid + SFXBT group. Evaluate the AR behavioral score, observe the pathological changes of nasal mucosa with H&E, detect the composition and content of colonic contents with 16S rDNA, the levels of serum IgE, IL-4 and short-chain fatty acid (SCFAs) with ELISA, the mRNA expressions of STAT5, STAT6 and GATA3 in nasal mucosa with RT-qPCR, and the protein expressions of IL-4, STAT5, STAT6 and GATA3 in nasal mucosa with Western Blot. ResultsCompared with the OVA group, the AR behavioral score in the antibiotic + SFXBT group and acetic acid + SFXBT group decreased (P < 0.01), and the pathological changes of nasal mucosa were alleviated. At the door level, the relative abundance of Firmicutes in feces increased significantly, while the relative abundance of Bacteroidetes decreased significantly. At the genus level, The relative abundance of fecal Lactobacillus, Romboutsia, Allobaculum and Dubosiella increased significantly, the levels of serum IgE and IL-4 decreased (P < 0.05), the content of SCFAs increased significantly (P < 0.05), and the expression levels of STAT5, STAT6 and GATA3 mRNA and protein in nasal mucosa decreased significantly (P < 0.05). ConclusionShufeng Xingbi Therapy can significantly improve the inflammatory symptoms of nasal mucosa in AR rats, and its mechanism may be closely related to regulating Th1/Th2 immune balance and intestinal flora.

immunology↗

bioRxiv2025

iGABASnFR2: Improved genetically encoded protein sensors of GABA

Monitoring GABAergic inhibition in the nervous system has been enabled by development of an intensiometric molecular sensor that directly detects GABA. However, the first generation iGABASnFR exhibits low signal-to-noise and suboptimal kinetics, making in vivo experiments challenging. To improve sensor performance, we targeted several sites in the protein for near-saturation mutagenesis and evaluated the resulting sensor variants in a high throughput screening system using evoked synaptic release in primary cultured neurons. This identified a sensor variant, iGABASnFR2, with 4.1-fold improved sensitivity and 30% faster rise time, and binding affinity that remained in a range sensitive to changes in GABA concentration at synapses. We also identified sensors with an inverted response, decreasing fluorescence intensity upon GABA binding. We termed the best such negative-going sensor iGABASnFR2n, which can be used to corroborate observations with the positive-going sensor. These improvements yielded a qualitative enhancement of in vivo performance when compared directly to the original sensor. iGABASnFR2 enabled the first measurements of direction-selective GABA release in the retina. In vivo imaging in somatosensory cortex revealed that iGABASnFR2 can report volume-transmitted GABA release following whisker stimulation. Overall, the improved sensitivity and kinetics of iGABASnFR2 make it a more effective tool for imaging GABAergic transmission in intact neural circuits.

neuroscience↗

bioRxiv2025

Glutamate indicators with increased sensitivity and tailored deactivation rates

Identifying the input-output operations of neurons requires measurements of synaptic transmission simultaneously at many of a neurons thousands of inputs in the intact brain. To facilitate this goal, we engineered and screened 3365 variants of the fluorescent protein glutamate indicator iGluSnFR3 in neuron culture, and selected variants in the mouse visual cortex. Two variants have high sensitivity, fast activation (< 2 ms) and deactivation times tailored for recording large populations of synapses (iGluSnFR4s, 153 ms) or rapid dynamics (iGluSnFR4f, 26 ms). By imaging action-potential evoked signals on axons and visually-evoked signals on dendritic spines, we show that iGluSnFR4s/4f primarily detect local synaptic glutamate with single-vesicle sensitivity. The indicators detect a wide range of naturalistic synaptic transmission, including in the vibrissal cortex layer 4 and in hippocampal CA1 dendrites. iGluSnFR4 increases the sensitivity and scale (4s) or speed (4f) of tracking information flow in neural networks in vivo.

neuroscience↗

bioRxiv2025

RPGR regulates motile cilia by interfering with actin dynamics

Cilia are highly conserved cellular organelles extruding from the surface of cell types carrying either sensory (signaling) or motile functions. These include photoreceptor cells and airway epithelia, where they function in light sensation and mucociliary clearance respectively. Retinitis pigmentosa GTPase regulator (RPGR) variants affect both photoreceptor sensory cilia and airway motile cilia, leading to retinitis pigmentosa (RP) and in some cases, primary ciliary dyskinesia (PCD), both debilitating conditions. Not all patients develop PCD and it is unclear which RPGR variants predispose patients to PCD and why this happens. In this study, using nasal biopsy samples of patients with RPGR-related RP, we leverage 2D organoid cell culturing, super-resolution microscopy, and live cell imaging to characterize the multiciliated cells from patients with different RPGR variants, healthy human nasal and bronchial multiciliated cells with CRISPR-modified RPGR function. We demonstrate for the first time that multiciliated cells with RPGR variants may have reduced ciliation, shorter cilia, significantly impaired cilia beat, or cilia beat incoordination, which could lead to defective mucociliary clearance and lung disease. In addition, we show the regulation of motile cilia by RPGR involves F-actin, as evidenced by temporarily reduced Gelsolin and undissolved condensed actin meshwork at the apical surface of RPGR-deficient multiciliated cells. In support, we show that the motile cilia defect can be ameliorated by treating with the actin polymerization inhibitor Latrunculin A. Though PCD was observed only in patients with variants that affect both main isoforms (RPGR1-19 and RPGRORF15), patients with variants affecting only RPGRORF15 also showed cilia and airway anomalies. Though all RPGR variants affected motile cilia in one way or another, RPGR loss of function variants affecting both isoforms are associated with more severe cilia and systemic phenotypes, the mechanisms of which involve the accumulation of apical F-actin. One Sentence Summary: Loss of RPGR, through the mechanism of apical F-actin accumulation in airway multiciliated cells, leads to reduced ciliation with short cilia that have an impaired beat, leading to defective mucociliary clearance.

cell biology↗

bioRxiv2025

GlycoGenius: the ultimate high-throughput glycan composition identification tool

Mass spectrometry is recognized as the gold standard for glycan analysis, yet the complexity of generated data hampers progress in glycobiology, as existing tools lack full automation, requiring extensive manual effort. We introduce GlycoGenius, an open-source program offering an automated workflow for glycomics data analysis, featuring an intuitive graphical interface. With algorithms tailored to reduce manual workload, it allows for data visualization and automatically constructs search spaces, identifies, scores, and quantifies glycans, filters results, and annotates fragment spectra of N- and O-glycans, glycosaminoglycans and more. It seamlessly guides researchers of all expertise levels from raw data to publication-ready figures. Our findings demonstrate that GlycoGenius achieves results comparable with manual analysis or competing tools, identifying more glycans, including novel ones, while significantly reducing processing time. This groundbreaking tool represents a significant advancement in the study of glycoconjugates, empowering researchers to focus on insights rather than data processing.

bioinformatics↗

bioRxiv2025

Yeast aging from a dynamic systems perspective: Analysis of single cell trajectories reveals significant interplay between nuclear size scaling, proteasome dynamics, and mitochondrial morphology

Yeast replicative aging is cell autonomous and thus a good model for mechanistic study from a dynamic systems perspective. Utilizing an engineered strain of yeast with a switchable genetic program to arrest daughter cells (without affecting mother cell divisions) and a high throughput microfluidic device, we systematically analyze the dynamic trajectories of thousands of single yeast mother cells throughout their lifespan, using fluorescent reporters that cover a range of biological processes, including some major aging hallmarks. We found that the markers of proteostasis stand out as most predictive of the lifespan of individual cells. In particular, nuclear proteasome concentration at middle age is a good predictor. We found that cell size (measured by area) grows linearly with time, and that nuclear size grows in proportion to maintain isometric scaling in young cells. As the cells become older, their nuclear size increases faster than linear and isometric size scaling breaks down. We observed that proteasome concentration in the nucleus exhibits dynamics very different from that in cytoplasm, with much more rapid decrease during aging; such dynamic behavior can be accounted for by the change of nuclear size in a simple mathematical model of transport. We hypothesize that the gradual increase of cell size and the associated nuclear size increase lead to the dilution of important nuclear factors (such as proteasome) that drives aging. We also show that perturbing proteasome changes mitochondria morphology and function, but not vice versa, potentially placing the change of proteosome upstream of the change of mitochondrial phenotypes. Our study produced large scale single cell dynamic data that can serve as a valuable resource for the aging research community to analyze the dynamics of other markers and potential causal relations between them. It is also a useful resource for building and testing physics/AI based models that identify early dynamics events predictive of lifespan and can be targets for longevity interventions.

bioinformatics↗

bioRxiv2025

DEK::NUP214 acts as an XPO1-dependent transcriptional activator of essential leukemia genes

The t(6;9)(p22.3;q34.1) translocation/DEK::NUP214 fusion protein defines a distinct subgroup of younger AML patients classified as a separate disease entity by the World Health Organization. DEK is a nuclear factor with multifunctional roles, including gene regulation, while its fusion partner, NUP214, plays a pivotal role in nuclear export by interacting with transport receptors such as XPO1. However, the precise mechanism by which DEK::NUP214 drives leukemia remains unclear. A comprehensive multi-omics comparison of 57 AML primary samples (including whole genome sequencing, targeted sequencing, transcriptomics, and drug screening with > 500 compounds) revealed that t(6;9) cases display a selective response to XPO1 inhibitors (Selinexor & Eltanexor) and a distinct transcriptomic signature characterized by the overexpression of FOXC1 and HOX genes that are key leukemia mediators. CUT&RUN experiments demonstrated the direct binding of DEK::NUP214 to the promoters of FOXC1 and HOXA/B clusters. Strikingly, the expression of these genes and the binding of DEK::NUP214 to their regulatory regions were selectively reduced upon XPO1 inhibition in t(6;9) cells. Altogether, these results identified a novel function of DEK::NUP214 as an XPO1-dependent transcriptional activator of key leukemia drivers and provide a rationale to explore the use of XPO1 inhibitors in this patient population.

cancer biology↗

bioRxiv2025

A distinctive subset of microglia positioned at the paraventricular zone are dedicated to cleaning the cerebrospinal fluid

The interconnected ventricles in the central nervous system are filled with cerebrospinal fluid (CSF) which serves multifaceted roles including bring away metabolic wastes. However, whether there is an inherent "dislodging" mechanism, in addition to CSF drainage and reabsorption, at play for the removal of microparticles in the CSF is unclear. In this study, we identified a subset of microglia distributed in close proximity to the ependymal wall of ventricles. They differed from parenchymal microglia in morphology, behaviors and transcriptomes. In particular, the paraventricular microglia extended transependymal dendrites into the lumen of ventricles and were proficient in sequestering and phagocytosing intraventricular exogenous particles when present. A specific removal of the paraventricular microglia led to an acute ventriculomegaly due to an increased colloid osmotic pressure in the CSF resulting from protein accumulation. Thus, we identified the paraventricular microglia as specific guardians in monitoring and cleaning the CSF.

immunology↗

bioRxiv2025

Graph-based Contrastive Learning Enables Unified Integration and Niche Transfer Across Single-Cell and Spatial Multi-Omics

The rapid growth of single-cell and spatial omics has outpaced computational methods capable of unifying these data into a cohesive framework for tissue atlas construction and cross-sample analysis. A critical bottleneck lies in the inability of existing tools to co-embed cells from diverse technologies--spanning transcriptomics, epigenomics, and proteomics--into a shared reference space while preserving spatial architecture and molecular specificity. Here, we present Garfield (Graph-based Contrastive Learning Enables Fast Single-Cell Embedding), a geometric deep-learning framework that addresses these challenges through spatially or molecularly aware cell embedding. Leveraging a graph contrastive learning framework, Garfield learns a shared embedding space for data generated by diverse technologies, enabling seamless construction and querying of spatial reference atlases. Our results show that Garfield consistently outperforms state-of-the-art benchmark models in identifying spatial niches across multiple datasets. We further demonstrate Garfields versatility by applying it to multi-modal spatial data, including gene expression and chromatin accessibility, where it successfully identifies distinct niches in the mouse brain. Notably, Garfield reveals tumor microenvironment heterogeneity in non-small cell lung cancer and breast cancer, uncovered conserved, barrier-like immune niches at tumor margins orchestrating CD80-mediated T cell-B cell-dendritic cell interactions and IFN-/B cell activation pathways, forming spatially coordinated immune surveillance hubs. These findings underscore Garfields potential to advance spatial omics research by offering a robust, scalable solution for integrating and interpreting complex spatial data across diverse tissue types and modalities.

bioinformatics↗

bioRxiv2025

A Critical Role for the Fascin Family of Actin Bundling Proteins in Axon Development, Brain Wiring and Function

Actin-based cell motility drives many neurodevelopmental events including guided axonal growth. Fascin is a major family of F-actin bundling proteins, but its role in axon development in vivo and brain wiring remains unclear. Here, we report that fascin is required for axon development, brain wiring and function. We show that fascin is enriched in the motile filopodia of axonal growth cones and its inhibition impairs axonal extension and branching of hippocampal neurons in culture. We next provide evidence that fascin is essential for axon development and brain wiring in vivo using Drosophila melanogaster as a model. Drosophila expresses a single ortholog of mammalian fascin called Singed (SN), which is expressed in the mushroom body (MB) of the central nervous system. Loss of SN causes severe MB disruption, marked by - and {beta}-lobe defects indicative of altered axonal guidance. SN-null flies also exhibit defective sensorimotor behaviors as assessed by the negative geotaxis assay. MB-specific expression of SN in SN-null flies rescues MB structure and sensorimotor deficits, indicating that SN functions autonomously in MB neurons. Together, our data from primary neuronal culture and in vivo models highlight a critical role for fascin in brain development and function. HighlightsO_LIFascin regulates axon growth and branching of hippocampal neurons in culture. C_LIO_LISinged, a Drosophila fascin ortholog, is enriched in mushroom body (MB) axons. C_LIO_LISinged loss causes axon guidance defects and sensorimotor issues in flies. C_LIO_LIMB-specific Singed re-expression rescues MB structure and behavior in flies. C_LI

cell biology↗

bioRxiv2025

Learning universal knowledge graph embedding for predicting biomedical pairwise interactions

Predicting biomedical interactions is crucial for understanding various biological processes and drug discovery. Graph neural networks (GNNs) are promising in identifying novel interactions when extensive labeled data are available. However, labeling biomedical interactions is often time-consuming and labor-intensive, resulting in low-data scenarios. Furthermore, distribution shifts between training and test data in real-world applications pose a challenge to the generalizability of GNN models. Recent studies suggest that pre-training GNN models with self-supervised learning on unlabeled data can enhance their performance in predicting biomedical interactions. Here, we propose LukePi, a novel self-supervised pre-training frame-work that pre-trains GNN models on biomedical knowledge graphs (BKGs). LukePi is trained with two self-supervised tasks: topology-based node degree classification and semantics-based edge recovery. The former is to predict the degree of a node from its topological context and the latter is to infer both type and existence of a candidate edge by learning semantic information in the BKG. By integrating the two complementary tasks, LukePi effectively captures the rich information from the BKG, thereby enhancing the quality of node representations. We evaluate the performance of LukePi on two critical link prediction tasks: predicting synthetic lethality and drug-target interactions, using four benchmark datasets. In both distribution-shift and low-data scenarios, LukePi significantly outperforms 15 baseline models, demonstrating the power of the graph pre-training strategy when labeled data are sparse.

bioinformatics↗

bioRxiv2025

Beyond RNA Structure Alone: Complex-Aware Feature Fusion for Tertiary Structure-based RNA Design

Tertiary structure-based RNA design plays a crucial role in synthetic biology and therapeutics. While existing methods have explored structure-to-sequence mappings, they focus solely on RNA structures and overlook the role of complex-level information, which is crucial for effective RNA design. To address this limitation, we propose tthe Complex-Aware tertiary structure-based RNA Design model, CARD, that integrates complex-level information to enhance tertiary structure-based RNA sequence design. To be specific, our method incorporates protein features extracted by protein language model (e.g., ESM-2), enabling the design model to generate more accurate and complex relevant sequences. Considering the biological complexity of protein-RNA interactions, we introduce a distance-aware filtering for local features from protein representation. Furthermore, we design a high-affinity design framework that combines our CARD with an affinity evaluation model. In this framework, candidate RNA sequences are generated and rigorously screened based on affinity and structural alignment to produce high-affinity RNA sequences. Extensive experiments demonstrate the effectiveness of our method with an improvement of 5.6% compared with base model without our complex-aware feature integration. A concrete case study for 2LBS further validates the superiority of our CARD.

bioinformatics↗