Biology subjects

Ye, R.

Publications and source records attributed to Ye, R..

At least 19 recordsLinked to original sources

bioRxiv2026

Language-guided domain adaptation improves protein sequence rep-resentations for antimicrobial resistance analysis

Antimicrobial resistance (AMR), particularly antibiotic resistance in bacterial pathogens, complicates infectious disease management. Because bacterial resistance is often mediated by protein functions, including antibiotic inactivation and efflux, protein sequence analysis provides a molecular basis for digital AMR surveillance and therapeutic intervention. We developed BacLLM, a language guided protein representation framework that integrates annotation-derived functional knowledge into protein language model representations for AMR analysis in WHO priority bacterial pathogens. Through sequence enhancement, semantic enrichment of UniProt annotations, and cross modal contrastive learning, BacLLM transfers LLM encoded annotation priors into the sequence encoder. Compared with baselines, BacLLM provided more informative representations for resistance-gene identification and mechanism prediction, captured resistance associated mutational signals at single-residue resolution, and preserved taxonomic relationships not reflected by phenotype based clinical groupings. Moreover, BacLLM recovered resistance associated functional organization within a TolC-centered protein association network, prioritizing proteins with low sequence identity to TolC but predicted structural similarity to TolC. Extending from AMR characterization to therapeutic prioritization, BacLLM also improved strain-level phage host matching for phage therapy candidate prioritization. Together, these findings establish BacLLM as a protein representation framework supporting AMR surveillance, functional characterization, and therapeutic candidate prioritization.

bioinformatics↗

bioRxiv2026

A community benchmark enables high-resolution, cross-lab quantification of mouse social behavior

Social behavior is a sensitive readout of brain state, but quantifying rodent social behavior at scale has relied on siloed, lab-specific tools. We assembled Days of our Mice (DooM), a benchmark of manually annotated video and pose tracking spanning 15 laboratories and 33 social and solitary behaviors, and hosted a community behavior classification challenge. We built on challenge solutions to develop the High-Dimensional Rodent Annotator (HiDRA), a behavior classification tool that generalizes to labs unseen during training and learns cross-lab movement features. Applied without retraining to experimental data, HiDRA resolves line-specific and state-dependent behavioral fingerprints with strong generalizability and reproducibility, establishing high-resolution, shareable social phenotyping as a scalable readout of brain and internal state. We release the DooM dataset and HiDRA classifiers as open resources.

neuroscience↗

bioRxiv2026

InfluProto: Predicting Cross-Species Transmission of Influenza A Virus via Prototype Learning

Zoonotic influenza A viruses pose constitute a significant persistent threat to global public health, yet quantitative approaches for assessing their cross-species transmission potential remain limited. Here, we present InfluProto, a deep learning model that introduces prototype learning into host prediction for influenza A virus, framing conceptualizing host adaptation as a continuous process rather than a discrete outcome. InfluProto embeds viral sequences into a continuous representation space anchored by human, swine, and avian prototypes, using cosine distances as a quantitative measure of host adaptation. InfluProto accurately discriminated among the three host classes, achieving recall and precision exceeding 0.97, and substantially outperformed conventional classification models models--particularly on challenging zoonotic strains, including the 2009 pandemic H1N1 (pdm09) lineage. Beyond host classification, the learned representations captured biologically meaningful host-adaptive signatures,; notably, with pdm09 viruses occupying occupied an intermediate position between human and swine clusters, consistent with their evolutionary origin. Analysis of simulated reassortment further revealed that the introduction of human-origin segments into swine influenza viruses reduced their distance to the human prototype, indicating that InfluProto can detect reassortment-associated shifts in host adaptation. To facilitate broad public useaccess, we have developed an interactive web service for InfluProto. Collectively, our work establishes a continuous framework for characterizing host adaptation and evaluating zoonotic risk in influenza A viruses, with potential implications for pandemic surveillance and preparedness.

microbiology↗

bioRxiv2026

Analysis of Rare Coding Variation Identifies New Genetic Contributors to Schizophrenia

In this study, we present the largest rare coding variant association study of schizophrenia to date, with 87,959 schizophrenia cases and 150,587 controls. We identify 16 genes at exome-wide significance: SETD1A, ZMYM2, HERC1, RB1CC1, SCAF1, XPO7, SP4, FYN, PPP3CA, CUL1, HDAC9, JARID2, ATP9A, PTK2, STAG1, and SCN2A, and an additional 24 at a 5% false discovery rate. Cases carrying ultra-rare, damaging variants, primarily protein-truncating and deleterious missense mutations, show strong enrichment in constrained genes with consistent effects across ancestries. Half of the 40 identified genes overlap with those implicated in developmental delay, autism, or bipolar disorder, underscoring shared neurodevelopmental risk. All 40 genes identified are highly expressed in excitatory and inhibitory neurons, and their expression patterns span diverse developmental stages. Functionally, the identified genes implicate chromatin regulation, protein degradation, and synaptic function. Overall, our results advance understanding of the genetic architecture of schizophrenia, provide a foundation for modeling risk genes in cellular and animal systems, and offer a framework for future work toward biologically informed patient stratification.

genetics↗

bioRxiv2026

Exploration of the screening and regulatory mechanisms of biomarkers related to ac4C modification in laryngeal squamous cell carcinoma patients based on single-cell analysis and machine learning

BackgroundN4-acetylcytidine (ac4C) modification plays a critical role in cancer development. Exploring ac4C modification in laryngeal squamous cell carcinoma (LSCC) may help elucidate its pathogenesis. MethodsLSCC-related datasets were obtained from GEO. After preprocessing and annotating single-cell data, malignant cells were identified by CNV scoring and further divided into subpopulations. Malignant epithelial cells (MECs) were identified and subclustered based on ac4C-related gene activity. Prognostic genes were screened using Cox regression and machine-learning approaches, followed by validation in clinical samples using qPCR. The biological and immunological relevance of these genes was further explored through immune infiltration, immunotherapy response, and mutation analyses. ResultsThe 14,465 identified MECs were classified into five subgroups (MEC1-5), among which MEC3 showed the strongest association with the ac4C gene set. Machine-learning analysis of MEC3-derived genes yielded seven prognostic markers, including BARX1, FHL2, NXPH4, PKMYT1, TNFAIP8L1, CRLF1, and CENPP. qPCR confirmed their differential expression between tumor and adjacent normal tissues. These genes were significantly associated with alterations in the tumor immune microenvironment, with high-risk patients showing increased immune infiltration and immune activity. ConclusionSeven ac4C-related prognostic genes were identified that may contribute to LSCC progression by modulating the tumor immune microenvironment, providing potential therapeutic insights.

developmental biology↗

bioRxiv2025

Global Mapping of circRNA-Target RNA Interactions Reveals P-Body-Mediated Translational Repression

Circular RNAs (circRNAs) are primarily produced through pre-mRNA back-splicing, yet their target repertoire and functional mechanisms remain elusive. Here, we present circTargetMap, a computational framework for globally mapping circRNA targets using RNA-RNA interactomes obtained via RNA in situ conformation sequencing (RIC-seq) in the hippocampus and ten human cell lines. This approach identified 117,163 high-confidence circRNA-target RNA interactions, with 83% of target mRNAs bound by multiple circRNAs. Functionally, CDR1as and circRMST repress target mRNA translation by sequestering them into processing bodies (P-bodies)--membraneless granules--through sequence-specific base-pairing, probably independent of AGO2, DICER, and miRNAs. To directly capture granule-associated interactions, we developed granule RIC-seq (GRIC-seq) method, revealing the broad role of circRNA-target RNA interactions in translational repression. Moreover, pathogenic variants are significantly enriched around circRNA-target RNA interaction sites, suggesting potential roles in disease. Our study provides valuable resources for circRNA functional exploration and a framework for investigating RNA-RNA interactions within membraneless organelles.

molecular biology↗

bioRxiv2025

Captamer: A Novel Quantitative Protein Detecting Method Depending on Aptamer-activated Molecular Switches and RPA Signal Amplification

There are various protein assays for specific and quantitative detection and widely used for laboratory and clinic purposes, but current methods still have limitations. Immunoassays based on antibodies, like ELISA, suffer from slow response and a long antibody-screening period, while physical or electrochemical methods are generally restricted by high cost or the stringent requirement of equipment or operating skills. In this study, we developed an in vitro sensitive protein quantification method: Captamer. The Captamer system comprises a molecular switch derived from aptamer sequence and an exponential fluorescence signal amplification pathway based on recombinase polymerase amplification (RPA). We demonstrated the Captamer for SARS-CoV-2 nucleocapsid protein detection and obtained results from samples within 30 min, displaying a wide detection window from 0.2 pg/mL to 200 pg/mL with high specificity. Furthermore, we tested the Captamer for Tau441 protein (a potential Alzheimers disease biomarker) and thrombin (a classic aptamer-protein interaction model), showing the limit of detection as low as 1 ng/mL and 0.02pg/mL respectively, which suggested the capacity of Captamer to be applied to various aptamer-protein pairs. Compared with the most commonly used and recently reported protein quantification methods, Captamer stands out for its high sensitivity, short response time, low cost, and simplicity, indicating its great potential to be widely used in protein quantification. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=131 SRC="FIGDIR/small/686180v1_ufig1.gif" ALT="Figure 1"> View larger version (32K): org.highwire.dtl.DTLVardef@1e38ffaorg.highwire.dtl.DTLVardef@1019e72org.highwire.dtl.DTLVardef@1498b68org.highwire.dtl.DTLVardef@15f5ea2_HPS_FORMAT_FIGEXP M_FIG C_FIG

biochemistry↗

bioRxiv2025

Influ-BERT: An Interpretable Model for Enhancing Low-Frequency Influenza A virus Subtype Recognition

Influenza A Virus (IAV) poses a persistent threat to global public health due to its broad host adaptability, frequent anti-genic variation, and potential for cross-species transmission. Accurate identification of IAV subtypes is essential for effective epidemic surveillance and precise disease control. Here, we present Influ-BERT, a domain-adaptive pretrained model based on the Transformer architecture. Optimized from DNABERT-2, Influ-BERT was developed using a dedicated corpus of approximately 900,000 influenza genome sequences. We constructed a custom Byte Pair Encoding (BPE) tokenizer, and employed a two-stage training strategy involving domain-adaptive pretraining followed by task-specific fine-tuning. This approach significantly enhanced identification performance for IAV subtypes. Experimental results demonstrate that Influ-BERT outperforms both traditional machine learning approaches and general genomic language models, such as DNABERT-2 and MegaDNA, in the task of IAV subtype identification. The model achieved F1-scores consistently above 97% and exhibited stable performance gains for subtypes that are underrepresented in sequencing data, including H5N8, H5N1, H7N9, and H9N2. Beyond subtype identification, Influ-BERT was successfully applied to additional tasks including respiratory virus identification, IAV pathogenicity prediction, and identification of IAV genomic fragments and functional genes, demonstrating robust performance throughout. Further interpretability analysis using sliding window perturbation confirmed that the model focuses on biologically significant genomic regions, providing insight into its improved predictive capability. Contactsongshh@big.ac.cn (Song S), atrv@lncc.br (Ana Tereza Ribeiro Vasconcelos)

bioinformatics↗

bioRxiv2025

Hippocampal input-driven plasticity of prefrontal interneurons reveals a circuit basis for impaired spatial working memory.

Long-range projections from ventral hippocampus (vHPC) to medial prefrontal cortex (mPFC) support cognitive functions including spatial working memory (SWM). vHPC input targets multiple prefrontal interneuron classes, yet how hippocampal input recruits these populations in vivo, whether this recruitment is plastic, and how such plasticity influences cognition remain unknown. Here, we combined optical stimulation of mouse vHPC inputs with calcium recordings from discrete mPFC interneuron populations, revealing persistent activity-induced reweighting of hippocampal recruitment across prefrontal inhibitory microcircuits. Ex vivo electrophysiology and computational modeling implicated weakened monosynaptic hippocampal drive onto vasoactive intestinal polypeptide (VIP)-expressing interneurons in this circuit reconfiguration. Prior vHPC input stimulation and the schizophrenia-associated Df(16)A+/- mutation, which also yielded reduced hippocampal input onto VIP interneurons, produced convergent alterations in task-related VIP interneuron activity associated with poorer SWM task learning. These findings implicate hippocampal input to VIP interneurons as a key locus of plasticity capable of reconfiguring inhibitory microcircuit activity linked to impaired working memory task learning. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=189 SRC="FIGDIR/small/665987v2_ufig1.gif" ALT="Figure 1"> View larger version (52K): org.highwire.dtl.DTLVardef@e26cb0org.highwire.dtl.DTLVardef@14d9d6borg.highwire.dtl.DTLVardef@9a0fc2org.highwire.dtl.DTLVardef@80afd1_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical AbstractC_FLOATNO C_FIG

neuroscience↗

bioRxiv2025

Single-cell alternative polyadenylation analysis reveals mechanistic insights of COVID-19-associated neurological and psychiatric effects

COVID-19 is associated with increased risks of neurological and psychiatric sequelae. Alternative polyadenylation (APA) is ubiquitous in human genes, resulting in mRNA diversity, and has been validated to play a pivotal regulatory role in the onset and progression of a variety of diseases, including viral infections. Here, we analyzed the APA usage across different cell types in frontal cortex cells from non-viral control group and COVID-19 patients, and identified functionally related APA events in COVID-19. According to our study, the poly(A) site (PAS) usage is different among cell types and following SARS-COV-2 infection. Moreover, we found the genes with significant PAS level changes affected pathways related to RNA splicing, and neuronal development and function, suggesting that survivors of COVID-19 will have a high risk of these diseases and that alternative splicing functions cause these changes. Additionally, APA usage and its correlation with gene expression levels varied across genes, some prefer short isoform that is more stable to produce more proteins, while others may be regulated by different mechanisms. A total of 267 risk genes targeted by microRNAs for common neurological and psychiatric disorders were found to undergo significant changes in APA following infection. In conclusion, our comprehensive analysis of APA in neural cells from COVID-19 patients at the single-cell level elucidated changes in APA levels in the brains of SARS-COV-2-infected patients and confirmed that these changes impair the function of the nervous system, providing important insights for COVID-19-associated sequelae. HighlightsO_LIWe analyzed the changes in single-cell APA level in the frontal cortex after SARS-COV-2 infection and found that PAS usage is different among cell types and changed after infection. C_LIO_LIWe find that APA changes are related to RNA splicing, and neuronal development and function, such as synaptic plasticity, nervous system development, learning or memory and cognition, and the brain function may be impaired due to global gene expression and APA changes. C_LIO_LIA total of 267 risk genes for common neurological and psychiatric disorders were found to undergo significant changes in APA following infection, implying that APA alterations in specific genes could be a potential therapeutic target and predictive biomarker for neurological and psychiatric disease. C_LIO_LIAPA in infection samples results in the removal/addition of miRNA target sites in the 3'-UTR of genes. A total of 2404 miRNAs were predicted which suggests that the miRNAs target site information can indicate the impact of APA events and has the potential to be used as a predictive biomarker. C_LI

bioinformatics↗

bioRxiv2025

Novel multi-transmitter calyx-like giant synapse revealed in the extended amygdala

We have discovered a highly specialized innervation of the forebrain by pituitary adenylate cyclase activating polypeptide (PACAP) immunohistochemistry originating from the brain stem that uses glutamate, acetylcholine and PACAP and other peptides as neurotransmitters. The parent neurons of the axons are in the Kolliker-Fuse nucleus and their terminals form calyx-like multi-release site synapses in the rodent forebrain extended amygdala similar to the calyx of Held in the auditory brain stem. The latter is a giant, excitatory, cup-like axo-somatic high-fidelity synapse. The PACAP-positive terminals also form enveloping axo-somatic specialization with mixed glutamatergic and cholinergic molecular identities, co-expressing VGluT1, VGluT2, VAChT, and the neuropeptides PACAP, CGRP, and neurotensin, together with calretinin in the presynaptic compartment. We identified a distinct neuronal subpopulation in the pontine Kolliker-Fuse region of the parabrachial complex that gives rise to these calyceal terminals, which engulf PKC/GluD1 somata in the capsular central amygdala and oval bed nucleus of the stria terminalis. Strikingly, GluD1 immunolabeling is concentrated at axo-somatic contact zones apposed to VAChT presynaptic vesicle clusters zones but is absent from postsynaptic densities of conventional type I synapses within the same terminals. The results demonstrate a previously unrecognized multimodal calyx-like synapse in the forebrain, with parallel fast ionotropic and modulatory peptidergic neurotransmission mechanisms, which is a substrate for high-fidelity signal transmission within viscerosensory -emotional circuits. Significance StatementA high-fidelity axo-somatic glutamatergic synapse is exemplified by the calyx of Held in the auditory brainstem, but is unknown in the forebrain. Here we reveal a multimodal calyceal synapse in the extended amygdala formed by glutamatergic afferents of the pontine Kolliker-Fuse nucleus that contain also several neuropeptides. These giant terminals co-package glutamate, acetylcholine, PACAP, CGRP and neurotensin. They envelop select PKC/GluD1 neurons in the capsular central amygdala and oval BNST. These unique forebrain synapses link viscerosensory input from the hindbrain to limbic autonomic centers. Such high-fidelity multimodal transmission is likely to underlie the precise synchronization of emotional and homeostatic control circuits.

neuroscience↗

bioRxiv2025

MKFGO: Integrating Multi-Source Knowledge Fusion with Pre-Trained Language Model for High-Accuracy Protein Function Prediction

Accurately identifying protein functions is essential to understand life mechanisms and thus advance drug discovery. Although biochemical experiments are the gold standard for determining protein functions, they are often time-consuming and labor-intensive. Here, we proposed a novel composite deep-learning method, MKFGO, to infer Gene Ontology (GO) attributes through integrating five complementary pipelines built on multi-source biological data. MKFGO was rigorously benchmarked on 1522 non-redundant proteins, demonstrating superior performance over 11 state-of-the-art function prediction methods. Comprehensive data analyses revealed that the major advantage of MKFGO lies in its two deep-learning components, HFRGO and PLMGO, which derive handcraft features and protein large language model (PLM)-based features, respectively, from protein sequences in different biological views, with effective knowledge fusion at the decision-level. HFRGO leverages an LSTM-attention network embedded with handcraft features, in which the triplet loss-based guilt-by-association strategy is designed to enhance the correlation between feature similarity and function similarity. PLMGO employs the PLM to capture feature embeddings with discriminative functional patterns from sequences. Meanwhile, another three components provide complementary insights for further improving prediction accuracy, driven by protein-protein interaction, GO term probability, and protein-coding gene sequence, respectively. The source codes and models of MKFGO are freely available at https://github.com/yiheng-zhu/MKFGO.

bioinformatics↗

bioRxiv2025

Single-cell spatial transcriptome reveals pathological features of human hippocampus with sclerosis

More than half of epileptic patients ultimately turned to intractable epilepsy. Mesial temporal lobe epilepsy with hippocampal sclerosis (MTLE-HS), the most common type of intractable epilepsies, whose pathological mechanism remains elusive. Here, using 42 human hippocampal samples from surgical donors of MTLE (32 with and 10 without HS) through single-cell resolution Stereo-seq and histologcial experiments, we revealed spatially pathological changes of gene expression and cell type composition with HS systematically. After precise parcellation of hippocampal subregions and differentially expressed gene (DEG) analysis between each region with and without sclerosis, we found Cornu Ammonis (CA) subregions with higher number of DEGs were vulnerable to sclerosis, especially CA1 and CA3. Within CA1, we found that CA1-superficial and proximal areas were more vulnerable than CA1-deep and distal areas in sclerosis. Meanwhile, after analyzing 350,795 segmented cells from Stereo-seq, we found dramatically increasing density of astrocyte accompanied with significantly decreasing density of excitatory neurons in CA1, especially superficial and proximal CA1, and CA3 in sclerosis. In these two vulnerable subregions, proliferative astrocyte (P_astrocyte) and reactive astrocyte (R_astrocyte) were found to be enriched whereas apoptotic subtype of astrocyte (A_astrocyte), related with apoptotic pathway, was mainly located in alveus, which strengthened cell communication with reactive microglia (R_microglia) in HS, revealing the novel pathological feature in our work. The pseudotime analysis indicated that CA excitatory neurons underwent synaptic impairment, energy dysfunction, aging, and finally losing cell identity until death through autophagy or apoptosis. Besides, we also found a resilient subtype, EX_CA2-4.3, highly expressed extracellular matrix related genes including PDYN, and was increasing the interaction of BDNF-NTRK, NFASC-CNTN1 to withstand the damage from sclerosis. Together, our study provided a reference of human hippocampus with and without HS caused by MTLE, and highlighted the potential pathological mechanism on molecular and cellular level of MTLE-HS.

neuroscience↗

bioRxiv2024

A blended genome and exome sequencing method captures genetic variation in an unbiased, high-quality, and cost-effective manner

We deployed the Blended Genome Exome (BGE), a DNA library blending approach that generates low pass whole genome (1-4x mean depth) and deep whole exome (30-40x mean depth) data in a single sequencing run. This technology is cost-effective, empowers most genomic discoveries possible with deep whole genome sequencing, and provides an unbiased method to capture the diversity of common SNP variation across the globe. To evaluate this new technology at scale, we applied BGE to sequence >53,000 samples from the Populations Underrepresented in Mental Illness Associations Studies (PUMAS) Project, which included participants across African, African American, and Latin American populations. We evaluated the accuracy of BGE imputed genotypes against raw genotype calls from the Illumina Global Screening Array. All PUMAS cohorts had R2 concordance [≥]95% among SNPs with MAF[≥]1%, and never fell below [≥]90% R2 for SNPs with MAF<1%. Furthermore, concordance rates among local ancestries within two recently admixed cohorts were consistent among SNPs with MAF[≥]1%, with only minor deviations in SNPs with MAF<1%. We also benchmarked the discovery capacity of BGE to access protein-coding copy number variants (CNVs) against deep whole genome data, finding that deletions and duplications spanning at least 3 exons had a positive predicted value of [~]90%. Our results demonstrate BGE scalability and efficacy in capturing SNPs, indels, and CNVs in the human genome at 28% of the cost of deep whole-genome sequencing. BGE is poised to enhance access to genomic testing and empower genomic discoveries, particularly in underrepresented populations.

genomics↗

bioRxiv2024

The landscape of gene loss and missense variation across the mammalian tree informs on gene essentiality

BackgroundThe degree of gene and sequence preservation across species provides valuable insights into the relative necessity of genes from the perspective of natural selection. Here, we developed novel interspecies metrics across 462 mammalian species, GISMO (Gene identity score of mammalian orthologs) and GISMO-mis (GISMO-missense), to quantify gene loss traversing millions of years of evolution. GISMO is a measure of gene loss across mammals weighed by evolutionary distance relative to humans, whereas GISMO-mis quantifies the ratio of missense to synonymous variants across mammalian species for a given gene. RationaleDespite large sample sizes, current human constraint metrics are still not well calibrated for short genes. Traversing over 100 million years of evolution across hundreds of mammals can identify the most essential genes and improve gene-disease association. Beyond human genetics, these metrics provide measures of gene constraint to further enable mammalian genetics research. ResultsOur analyses showed that both metrics are strongly correlated with measures of human gene constraint for loss-of-function, missense, and copy number dosage derived from upwards of a million human samples, which highlight the power of interspecies constraint. Importantly, neither GISMO nor GISMO-mis are strongly correlated with coding sequence length. Therefore both metrics can identify novel constrained genes that were too small for existing human constraint metrics to capture. We also found that GISMO scores capture rare variant association signals across a range of phenotypes associated with decreased fecundity, such as schizophrenia, autism, and neurodevelopmental disorders. Moreover, common variant heritability of disease traits are highly enriched in the most constrained deciles of both metrics, further underscoring the biological relevance of these metrics in identifying functionally important genes. We further showed that both scores have the lowest duplication and deletion rate in the most constrained deciles for copy number variants in the UK Biobank, suggesting that it may be an important metric for dosage sensitivity. We additionally demonstrate that GISMO can improve prioritization of recessive disorder genes and captures homozygous selection. ConclusionsOverall, we demonstrate that the most constrained genes for gene loss and missense variation capture the largest fraction of heritability, GISMO can help prioritize recessive disorder genes, and identify the most conserved genes across the mammalian tree. O_FIG O_LINKSMALLFIG WIDTH=168 HEIGHT=200 SRC="FIGDIR/small/594531v1_ufig1.gif" ALT="Figure 1"> View larger version (79K): org.highwire.dtl.DTLVardef@11c3e46org.highwire.dtl.DTLVardef@13c6e70org.highwire.dtl.DTLVardef@76a6c7org.highwire.dtl.DTLVardef@1bf8d86_HPS_FORMAT_FIGEXP M_FIG C_FIG

genetics↗

bioRxiv2024

ProLLM: Protein Chain-of-Thoughts Enhanced LLM for Protein-Protein Interaction Prediction

The prediction of protein-protein interactions (PPIs) is crucial for understanding biological functions and diseases. Previous machine learning approaches to PPI prediction mainly focus on direct physical interactions, ignoring the broader context of nonphysical connections through intermediate proteins, thus limiting their effectiveness. The emergence of Large Language Models (LLMs) provides a new opportunity for addressing this complex biological challenge. By transforming structured data into natural language prompts, we can map the relationships between proteins into texts. This approach allows LLMs to identify indirect connections between proteins, tracing the path from upstream to downstream. Therefore, we propose a novel framework ProLLM that employs an LLM tailored for PPI for the first time. Specifically, we propose Protein Chain of Thought (ProCoT), which replicates the biological mechanism of signaling pathways as natural language prompts. ProCoT considers a signaling pathway as a protein reasoning process, which starts from upstream proteins and passes through several intermediate proteins to transmit biological signals to downstream proteins. Thus, we can use ProCoT to predict the interaction between upstream proteins and downstream proteins. The training of ProLLM employs the ProCoT format, which enhances the models understanding of complex biological problems. In addition to ProCoT, this paper also contributes to the exploration of embedding replacement of protein sites in natural language prompts, and instruction fine-tuning in protein knowledge datasets. We demonstrate the efficacy of ProLLM through rigorous validation against benchmark datasets, showing significant improvement over existing methods in terms of prediction accuracy and generalizability. Our results highlight the potential of LLMs to transform the field of PPI, serving as a robust potential tool for various categories of biological and medical research. The code is available at: https://github.com/MingyuJ666/ProLLM.

bioinformatics↗

bioRxiv2024

The landscape of regional missense mutational intolerance quantified from 125,748 exomes

Missense variants can have a range of functional impacts depending on factors such as the specific amino acid substitution and location within the gene. To interpret their deleteriousness, studies have sought to identify regions within genes that are specifically intolerant of missense variation. Here, we leverage the patterns of rare missense variation in 730,947 exome sequenced individuals in the Genome Aggregation Database (gnomAD v4.1.1) against a null mutational model to identify transcripts with regional differences in missense constraint. Missense-depleted regions are enriched for ClinVar pathogenic variants, de novo missense variants from individuals with neurodevelopmental disorders, and complex trait heritability. Following ClinGen calibration recommendations for the ACMG/AMP variant classification guidelines, we establish that variants within regions with <36% of their expected missense variation achieve moderate support for pathogenicity. We integrate this regional constraint measure into a missense deleteriousness metric (named MPC) that effectively stratifies rare and de novo missense variants in individuals with early-onset developmental conditions from controls. These results provide additional tools to aid in missense variant interpretation.

genomics↗

bioRxiv2024

Structural insights into agonist binding and activation of succinate receptor 1

Succinate is an intermediate of the citric acid cycle and serves important functions in energy homeostasis and metabolic regulation. Extracellular accumulation of succinate acts as a stress-induced signal through its G protein-coupled receptor, SUCNR1. Research on succinate signaling is hampered by the lack of high-resolution structures of the agonist-bound receptor. Here we present cryo-EM structures of SUCNR1-Gi complexes with the receptor bound to succinate and its non-metabolite derivative epoxysuccinate. Structural analysis of SUCNR1 identified key determinants for recognition of the dicarboxylate agonists in cis conformation. R2817.39 and Y832.64 are critical to ligand binding, but Y301.39 and R993.29 also participate in binding of succinate and epoxysuccinate, respectively. The extracellular loop 2, through F175ECL2 in its {beta}-hairpin, forms a hydrogen bond with one of the carboxyl groups and serves as a lid to cap the binding pocket for succinate. At the receptor-Gi protein interface, agonist binding induces the rearrangement of a hydrophobic network on TM5 and TM6, leading to transmembrane signaling through TM3 and TM7. The agonist-bound SUCNR1 structures shed light on molecular recognition of succinate for receptor signaling, that may promote further development of novel agonists, antagonists and biased agonists targeting SUCNR1.

biochemistry↗