Biology subjects

Ying, R.

Publications and source records attributed to Ying, R..

12 recordsLinked to original sources

bioRxiv2025

spRefine Denoises and Imputes Spatial Transcriptomics with a Reference-Free Framework Powered by Genomic Language Model

The analysis of spatial transcriptomics is hindered by high noise levels and missing gene measurements, challenges that are further compounded by the higher cost of spatial data compared to traditional single-cell data. To overcome this challenge, we introduce spRefine, a deep learning framework that leverages genomic language models to jointly denoise and impute spatial transcriptomic data. Our results demonstrate that spRefine yields more robust celland spot-level representations after denoising and imputation, substantially improving data integration. In addition, spRefine serves as a strong framework for model pre-training and the discovery of novel biological signals, as highlighted by multiple downstream applications across datasets of varying scales. Notably, spRefine enhances the accuracy of spatial ageing clock estimations and uncovers new aging-related relationships associated with key biological processes, such as neuronal function loss, which offers new insights for analyzing ageing effect with spatial transcriptomics.

genomics↗

bioRxiv2025

STPath: A Generative Foundation Model for Integrating Spatial Transcriptomics and Whole Slide Images

Spatial transcriptomics (ST) has shown remarkable promise in pathology applications, shedding light on the spatial organization of gene expression and its relationship to the tumor microenvironment. However, its clinical adoption remains constrained due to the limited scalability of current sequencing technologies. While recent methods attempt to infer ST from whole slide images (WSIs) using pretrained image encoders, they remain restricted by limited gene coverage, organ-specific training, and the need for dataset-specific fine-tuning. In light of this, we introduce STPath, a generative foundation model pretrained on a large-scale collection of WSIs paired with ST profiles. This extensive pretraining enables STPath to directly predict gene expression across 38,984 genes and 17 organs without requiring downstream fine-tuning. STPath integrates multiple data modalities, including histological images, gene expressions, organ type, and sequencing technology information, within a novel geometry-aware Transformer architecture. Unlike previous methods that directly map WSIs to gene expression, STPath is trained using a masked gene expression prediction objective guided by tailored noise schedules, effectively balancing between capturing gene-gene dependencies and performing high-quality predictions. We evaluate STPath across 6 tasks spanning 23 datasets and 14 biomarkers, including gene expression prediction, spot imputation, spatial clustering, biomarker prediction, gene mutation prediction, and survival prediction. These results demonstrate STPaths strong ability to infer spatially resolved gene expression and reveal crucial pathological structures within tissue samples, underscoring its promise for scalable ST-based pathology applications.

pathology↗

bioRxiv2025

Pre-training Genomic Language Model with Variants for Better Modeling Functional Genomics

Sequence-to-function models can predict gene expression from sequence data and be used to link genetic information with transcriptomics data to understand regulatory processes and their effects on complex phenotypes. The genomic language models are pre-trained with large-scale DNA sequences and can generate robust representations of these sequences by learning the genomic context. How-ever, few studies can estimate the predictability of gene expression levels and bridge these two classes of models together to explore individualized gene expression prediction. In this manuscript, we propose UKBioBERT as a DNA language model pre-trained with genetic variants from UK BioBank. We demonstrate that UKBioBERT generates informative embeddings capable of identifying gene functions, and improving gene expression prediction in cell lines, thereby enhancing our understanding of gene expression predictability. Building upon these embeddings, we combine UKBioBERT with state-of-the-art sequence-to-function architectures, Enformer and Borzoi, to create UKBioFormer and UKBioZoi. These models exhibit better performance in predicting highly predictable gene expression levels and can be generalized across different cohorts. Furthermore, UKBioFormer effectively captures the relationship between genetic variants and expression variations, enabling in-silico mutation analyses and eQTL identification. Collectively, our findings underscore the value of integrating genomic language models and sequence-to-function approaches for advancing functional genomics research.

bioinformatics↗

bioRxiv2025

UNICORN: Towards Universal Cellular Expression Prediction with an Explainable Multi-Task Learning Framework

Sequence-to-function analysis is a challenging task in human genetics, especially in predicting cell-type-specific multi-omic phenotypes from biological sequences such as individualized gene expression. Here, we present UNICORN, a new method with improved prediction performances than the existing methods. UNICORN takes the embeddings from biological sequences as well as external knowledge from pre-trained foundation models as inputs and optimizes the predictor with carefully-designed loss functions. We demonstrate that UNICORN outperforms the existing methods in both gene expression prediction and multiomic phenotype prediction at the cellular level and the cell-type level, and it can also generate uncertainty scores of the predictions. Moreover, UNICORN is able to link personalized gene expression profiles with corresponding genome information. Finally, we show that UNICORN is capable of characterizing complex biological systems for different disease states or perturbations. Overall, embeddings from foundation models can facilitate the understanding of the role of biological sequences in the prediction task, and incorporating multi-omic information can enhance prediction performances.

bioinformatics↗

bioRxiv2025

spEMO: Exploring the Capacity of Foundation Models for Analyzing Spatial Multi-Omic Data

Recent advances in pathology foundation models (PFMs), which are pretrained on large-scale histopathological images, have significantly accelerated progress in disease-centered applications. In parallel, spatial multi-omic technologies collect gene and protein expression levels at high spatial resolution, offering rich understanding of tissue context. However, current models fall short in effectively integrating these complementary data modalities. To fill in this gap, we introduce spEMO, a novel computational system that unifies embeddings from pathology foundation models and large language models (LLMs) to analyze spatial multi-omic data. By incorporating multimodal representations and information, spEMO outperforms models trained on single-modality data across a broad range of downstream tasks, including spatial domain identification, spot-type classification, whole-slide disease-state prediction and interpretation, inference of multicellular interactions, and automated medical report generation. The outstanding performances of spEMO in these tasks demonstrate its strength in both biological and clinical applications. Additionally, we propose a new evaluation task, known as multi-modal alignment, to assess the information retrieval capabilities of pathology foundation models. This task provides a principled benchmark for evaluating and improving model architectures. Collectively, spEMO represents a step forward in building holistic, interpretable, and generalizable AI systems for spatial biology and pathology.

bioinformatics↗

bioRxiv2024

ImmunoStruct: Integration of protein sequence, structure, and biochemical properties for immunogenicity prediction and interpretation

Epitope-based vaccines are promising therapeutic modalities for infectious diseases and cancer, but identifying immunogenic epitopes is challenging. The vast majority of prediction methods only use amino acid sequence information, and do not incorporate wide-scale structure data and biochemical properties across each peptide-MHC. We present ImmunoStruct, a deep-learning model that integrates sequence, structural, and biochemical information to predict multi-allele class-I peptide-MHC immunogenicity. By leveraging a multimodal dataset of[~] 27,000 peptide-MHCs, we demonstrate that ImmunoStruct improves immunogenicity prediction performance and interpretability beyond existing methods, across infectious disease epitopes and cancer neoepitopes. We further show strong alignment with in vitro assay results for a set of SARS-CoV-2 epitopes, as well as strong performance in peptide-MHC-based cancer patient survival prediction. Overall, this work also presents a new architecture that incorporates equivariant graph processing and multimodal data integration for the long standing task in immunotherapy.

immunology↗

bioRxiv2024

Chromatin interaction maps of human arterioles reveal new mechanisms for the genetic regulation of blood pressure

Arterioles are small blood vessels located just upstream of capillaries in nearly all tissues. The constriction and dilation of arterioles regulate tissue perfusion and are primary determinants of systemic blood pressure (BP). Abnormalities in arterioles are central to the development of major diseases such as hypertension, stroke, and microvascular complications of diabetes. Despite the broad and essential role of arterioles in physiology and disease, current knowledge of the functional genomics of arterioles is largely absent, partly because it is challenging to obtain and analyze human arteriole samples. Here, we report extensive maps of chromatin interactions, single-cell expression, and other molecular features in human arterioles and uncover new mechanisms linking human genetic variants to gene expression in vascular cells and the development of hypertension. Compared to large arteries, arterioles exhibited a higher proportion of pericytes which were strongly associated with BP traits. BP-associated single nucleotide polymorphisms (SNPs) were enriched in chromatin interaction regions in arterioles, particularly through enhancer SNP-promoter interactions, which were further linked to gene expression specificity across tissue components and cell types. Using genomic editing in animal models and human induced pluripotent stem cells, we discovered novel mechanisms linking BP-associated noncoding SNP rs1882961 to gene expression through long-range chromatin contacts and revealed remarkable effects of a 4-bp noncoding genomic segment on hypertension in vivo. We anticipate that our rich data and findings will advance the study of the numerous diseases involving arterioles. Moreover, our approach of integrating chromatin interaction mapping in trait-relevant tissues with SNP analysis and in vivo and in vitro genome editing can be applied broadly to bridge the critical gap between genetic discoveries and physiological understanding.

genomics↗

bioRxiv2024

Neural correlates of flexible sound perception in the auditory midbrain and thalamus

Hearing is an active process in which listeners must detect and identify sounds, segregate and discriminate stimulus features, and extract their behavioral relevance. Adaptive changes in sound detection can emerge rapidly, during sudden shifts in acoustic or environmental context, or more slowly as a result of practice. Although we know that context- and learning-dependent changes in the spectral and temporal sensitivity of auditory cortical neurons support many aspects of flexible listening, the contribution of subcortical auditory regions to this process is less understood. Here, we recorded single- and multi-unit activity from the central nucleus of the inferior colliculus (ICC) and the ventral subdivision of the medial geniculate nucleus (MGV) of Mongolian gerbils under two different behavioral contexts: as animals performed an amplitude modulation (AM) detection task and as they were passively exposed to AM sounds. Using a signal detection framework to estimate neurometric sensitivity, we found that neural thresholds in both regions improved during task performance, and this improvement was driven by changes in firing rate rather than phase locking. We also found that ICC and MGV neurometric thresholds improved and correlated with behavioral performance as animals learn to detect small AM depths during a multi-day perceptual training paradigm. Finally, we reveal that in the MGV, but not the ICC, context-dependent enhancements in AM sensitivity grow stronger during perceptual training, mirroring prior observations in the auditory cortex. Together, our results suggest that the auditory midbrain and thalamus contribute to flexible sound processing and perception over rapid and slow timescales. Significance statementWhat a listener hears depends on several factors, such as whether the listener is attentive or distracted, and whether the sound is meaningful or irrelevant. Practice can also shape hearing by improving the detection of particular sound features, as occurs during language or musical learning. Understanding how changes in sound perception are implemented in the brain is important for developing strategies to optimize healthy hearing, and for treating disorders in which these processes go awry. We report that neurons in auditory midbrain and thalamus exhibit rapid shifts in sound sensitivity that depend on the sounds behavioral relevance, and slower improvements that emerge over several days of training. Our results suggest that subcortical areas make an important contribution to flexible hearing.

neuroscience↗

bioRxiv2023

Organization of orbitofrontal-auditory pathways in the Mongolian gerbil

Sound perception is highly malleable, rapidly adjusting to the acoustic environment and behavioral demands. This flexibility is the result of ongoing changes in auditory cortical activity driven by fluctuations in attention, arousal, or prior expectations. Recent work suggests that the orbitofrontal cortex (OFC) may mediate some of these rapid changes, but the anatomical connections between the OFC and the auditory system are not well-characterized. Here, we used virally-mediated fluorescent tracers to map the projection from OFC to the auditory midbrain, thalamus, and cortex in a classic animal model for auditory research, the Mongolian gerbil (Meriones unguiculatus). We observed no connectivity between the OFC and the auditory midbrain, and an extremely sparse connection between the dorsolateral OFC and higher-order auditory thalamic regions. In contrast, we observed a robust connection between the ventral and medial subdivisions of the OFC and the auditory cortex, with a clear bias for secondary auditory cortical regions. OFC axon terminals were found in all auditory cortical lamina but were significantly more concentrated in the infragranular layers. Tissue-clearing and lightsheet microscopy further revealed that auditory cortical-projecting OFC neurons send extensive axon collaterals throughout the brain, targeting both sensory and non-sensory regions involved in learning, decision-making, and memory. These findings provide a more detailed map of orbitofrontal-auditory connections and shed light on the possible role of the OFC in supporting auditory cognition.

neuroscience↗

bioRxiv2022

Mutagenesis and structural modeling implicate RME-8 IWN domains as conformational control points

After endocytosis, transmembrane cargo is differentially sorted into degradative or recycling pathways. This process is facilitated by recruitment into physically distinct degradative or recycling microdomains on the limiting membrane of individual endosomes. Endosomal sorting complexes required for transport (ESCRT) mark the degradative microdomain, while the recycling domain is marked by the retromer complex and associated proteins RME-8 and SNX-1. The separation of endosomal microdomains is also controlled by RME-8 and SNX-1, at least in part via removal of degradative component HRS/HGRS-1 from the recycling microdomain. This activity is likely due to recruitment and activation of chaperone Hsc70 to the recycling microdomain by the RME-8 DNAJ domain. To better understand the mechanism of RME-8 function we performed a new phylogenetic analysis of RME-8 and identified new conserved sequence features. In a complementary approach, we performed structure-function analysis that identified the C-terminus as important for microdomain localization and likely substrate binding, while N-terminal sequences beyond the known single N-terminal PH-like domain are important for endosome recruitment. Random mutagenesis identified IWN4, and by analogy IWN3, to be important for the inhibitory DNAJ domain binding. Combining AlphaFold structural predictions with in vivo truncation and point mutation analysis of RME-8, we propose a model whereby SNX-1 and the IWN domains control the conformation of RME-8 and hence the productive exposure of the DNAJ domain. Furthermore, we propose that the activation of RME-8 is cyclical, with SNX-1 acting as an activator and a target of RME-8 uncoating activity.

cell biology↗

bioRxiv2021

A chicken IgY can efficiently inhibit the entry and replication of SARS-CoV-2 by targeting the ACE2 binding domain in vitro

COVID-19 pneumonia has now spread widely in the world. Currently, no specific antiviral drugs are available. The vaccine is the most effective way to control the epidemic. Passive immune antibodies are also an effective method to prevent and cure COVID-19 pneumonia. We used the SARS-CoV-2 S receptor-binding domain (RBD) as an antigen to immunize layers in order to extract, separate, and purify SARS-CoV-2-IgY from egg yolk. SARS-CoV-2-IgY (S-IgY)can block the entry of SARS-CoV-2 into the Cells and reduce the viral load in cells. The Half effective concentration (EC50) of W3-IgY (S-IgY in the third week after immunization) is 1.35 {+/-} 0.15nM. The EC50 of W9-IgY (S-IgY in the ninth week after immunization) is 2.76 {+/-} 1.54 nM. When the dose of S-IgY is 55 nM, the fluorescence representing intracellular viral protein is obviously weakened in Immunofluorescence microscopy. Results of Sars-CoV-2 /Vero E6 cell experiment confirmed that S-IgY has a strong antiviral effect on SARS-CoV-2, and its (EC50) is 27.78 {+/-}1.54 nMvs 3,259 {+/-} 159.62 nM of Redesivir (differ > 106 times P<0.001). S-IgY can inhibit the entry and replication of SARS-CoV-2, which is related to its targeting the ACE2 binding domain. S-IgY is safe, efficient, stable, and easy to obtain. This antibody can be an effective tool for preventing and treating COVID-19 pneumonia. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=143 SRC="FIGDIR/small/430255v3_fig1.gif" ALT="Figure 1"> View larger version (38K): org.highwire.dtl.DTLVardef@f24996org.highwire.dtl.DTLVardef@bd26b9org.highwire.dtl.DTLVardef@3956bborg.highwire.dtl.DTLVardef@6d4785_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOFig. 1.C_FLOATNO Graphical Abstract The figure briefly illustrates that the preparation and extraction of S-IgY and its anti-S-CoV-2 mechanism is to inhibit the entry and replication of SARS-CoV-2 by targeting the ACE2. C_FIG

immunology↗

bioRxiv2016

Modeling HIV disease progression and transmission at population-level: The potential impact of modifying disease progression in HIV treatment programs

IntroductionMathematical models of HIV transmission that incorporate the dynamics of disease progression can estimate the potential impact of adjunctive strategies to antiretroviral therapy (ART) for HIV treatment and prevention. Suppressive treatment of HIV-positive persons co-infected with herpes simplex virus-2 (HSV-2) with valacyclovir, a medication directed against HSV-2, can lower HIV viral load, but the impact of valacyclovir on population HIV transmission has not been estimated.\n\nMethodsWe applied data on CD4 and viral load progression in ART-naive persons studied in two HIV clinical trials to a novel, discrete-time Markov model. We validated our disease progression estimates using data from a trial of home-based HIV counseling and testing in KwaZulu-Natal, South Africa. Finally, we applied our disease progression estimates to a dynamic transmission model estimating the impact of providing valacyclovir to ART-naive individuals to reduce onward transmission of HIV in three scenarios of different ART and valacyclovir population coverage. We assumed that valacyclovir reduced HIV viral load by 1.23 log copies/L, and that persons treated with valacyclovir initiated ART more rapidly when their CD4 fell below 500 due to improved retention in pre-ART care.\n\nResultsThe average duration of HIV infection following acute infection was 9.5 years. The duration of disease after acute infection and before reaching CD4 200 cells/L was 2.53 years longer for females than males. Relative to a baseline of community HIV testing and counseling and ART initiation at CD4 <=500 cells/L, valacyclovir with increased linkage to care resulted in 166,000 fewer HIV infections over ten years, with an incremental cost-effectiveness ratio (ICER) of $4,696 per HIV infection averted. The Test and Treat scenario with 70% ART coverage and no valacyclovir resulted in 202,000 fewer HIV infections at an ICER of $6,579.\n\nConclusionEven when compared with initiation of valacyclovir, a safe drug that reduces HIV viral load, universal treatment for HIV is the optimal strategy for averting new infections and increasing public health benefit. Universal HIV treatment should be pursued by all countries to most effectively and efficiently reduce the HIV burden.

epidemiology↗