Biology subjects

Liu, D.

Publications and source records attributed to Liu, D..

At least 109 records · Page 6Linked to original sources

bioRxiv2025

YAP1 drives aggressive and therapy resistant state in melanoma through reprogramming the chromatin and regulating immune evasive programs

Despite promising initial results in targeting the RAF-MEK-ERK cascade, resistance to BRAF/MEK inhibitors remains a critical challenge in nearly 50% of melanoma patients. Our study demonstrates that robust YAP1 activation in metastatic melanoma correlates with poor survival and drives transcriptional programs linked to therapeutic resistance. Mechanistically, YAP1 predominantly remodels the chromatin landscape in resistant tumors by partnering with BRD4 and TEAD, creating a permissive transcriptional state that sustains oncogenic signaling. Clinical validation in biopsies from resistant melanoma confirms elevated expression of YAP1 target genes. Furthermore, pharmacological inhibition of BRD4 or TEAD reduces YAP1-driven transcription and reactivates antitumor immunity programs. TEAD specific inhibitors (and not verteporfin which is a highly non-specific inhibitor) synergize with immune checkpoint blockade in in vivo model system by promoting increased CD8 T cell infiltration and prolonged survival in the melanoma mouse model. Collectively, these findings reveal a chromatin-centric vulnerability in BRAF/MEK inhibitor-resistant melanoma and propose TEAD specific inhibitors as a promising dual strategy to overcome resistance and reinvigorate the immune response, offering a novel therapeutic avenue for patients. Resistance to BRAF and MEK inhibitors remains a major obstacle in the treatment of melanoma. Here, we show that elevated YAP1 activity in metastatic melanoma is associated with poor patient survival and drives transcriptional programs that promote therapeutic resistance. Mechanistically, YAP1 cooperates with BRD4 and TEAD to reprogram the chromatin landscape, establishing an oncogenic transcriptional state that sustains resistance. Analysis of patient-derived melanoma biopsies confirms increased expression of YAP1 target genes in resistant tumors. Targeting this epigenetic circuitry using pharmacological inhibitors of BRD4 or TEAD effectively suppresses YAP1-driven transcriptional output and restores antitumor immune activity, as evidenced by enhanced CD8 T cell infiltration and prolonged survival in resistant melanoma models. These effects are further supported by transcriptomic and chromatin accessibility analyses, which reveal reduced expression of immune-suppressive and proliferation-associated gene networks upon TEAD inhibition. Collectively, our findings identify a chromatin-centric mechanism underlying resistance to MAPK-targeted therapy and nominate TEAD inhibition as a promising dual-action strategy to both overcome resistance and re-engage antitumor immunity. This work offers a compelling therapeutic avenue for patients with drug-resistant melanoma.

cancer biology↗

bioRxiv2025

Towards an AI-based knowledge assistant for goat farmers based on Retrieval-Augmented Generation

Large language models (LLMs) are increasingly being recognised as valuable knowledge communication tools in many industries. However, their application in livestock farming remains limited, being constrained by several factors not least the availability, diversity and complexity of knowledge sources. This study introduces an intelligent knowledge assistant system designed to support health management in farmed goats. Leveraging the Retrieval-Augmented Generation (RAG), two structured knowledge processing methods - table textualization and decision-tree textualization - were proposed to enhance large language models (LLMs) understanding of heterogeneous data formats. Based on these methods, a domain-specific goat farming knowledge base was established to improve LLMs capacity for cross-scenario generalization. The knowledge base spans five key domains: Disease Prevention and Treatment, Nutrition Management, Rearing Management, Goat Milk Management, and Basic Farming Knowledge. Additionally, an online search module is integrated to enable real-time retrieval of up-to-date information. To evaluate system performance, six ablation experiments were conducted to examine the contribution of each component. The results demonstrated that heterogeneous knowledge fusion method achieved the best results, with mean accuracies of 87.90% on the validation set and 84.22% on the test set. Across the text-based, table-based, decision-tree based Q&A tasks, accuracy consistently exceeded 85%, validating the effectiveness of structured knowledge fusion within a modular design. Error analysis identified omission as the predominant error category, highlighting opportunities to further improve retrieval coverage and context integration. In conclusion, the results highlight the robustness and reliability of the proposed system for practical applications in goat farming.

bioengineering↗

bioRxiv2025

Claustrum-cortical reciprocal connections orchestrate allostatic responses following stress

Anxiety is an evolutionarily conserved response that supports threat detection and survival, yet dysregulated stress responses can drive its progression into pathological states. Here, we identify the claustrum (CLA) as a central hub that links acute stress responses to chronic affective dysfunction through cell-type-specific circuit dynamics. Acute social defeat stress robustly activated the CLA and induced anxiety-like behaviors in mice. Multimodal profiling revealed transcriptional and functional plasticity within CLA neurons, with Egr2-expressing excitatory and Gad2-expressing inhibitory populations exhibiting distinct adaptations to acute versus chronic stress. We further delineate a reciprocal CLA-anterior cingulate cortex (ACC) circuit that bidirectionally regulates stress responses in which glutamatergic CLAEgr2 [->] ACC projections promoted anxiety-like behaviors, whereas top-down ACC [->] CLAGad2inhibitory inputs constrained claustral output. Chronic stress induced hyperactivity in CLAGad2 inhibitory neurons and reshaped excitatory neuron dynamics, leading to increased intrinsic excitability, local circuit imbalance, and ultimately, depression-like phenotypes. These findings define a dynamic claustrum-centered circuit that governs the transition from adaptive anxiety to maladaptive affective states through balanced Egr2-Gad2 activity, and show that chronic-stress driven disruption of this excitatory-inhibitory equilibrium impairs stress coping and drives pathological behavioral outcomes. Graphic summaryAcute stress engages a claustrum (CLA)-centered network with activation of Egr2 excitatory neurons, promoting anxiety-like behaviors. Chronic stress drives CLAG142 neurons into a hyperactive state, enhancing local inhibition and suppressing excitatory output to induce depressive-like behaviors. A reciprocal CLAE7182[->]ACC feedforward and ACC[->]CLAG142 feedback circuit forms a state-dependent regulatory hub governing stress adaptation. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=180 SRC="FIGDIR/small/673892v2_ufig1.gif" ALT="Figure 1"> View larger version (31K): org.highwire.dtl.DTLVardef@75d664org.highwire.dtl.DTLVardef@4e5c31org.highwire.dtl.DTLVardef@86b5f2org.highwire.dtl.DTLVardef@135045_HPS_FORMAT_FIGEXP M_FIG C_FIG

neuroscience↗

bioRxiv2025

Distinct contributions and sensorimotor encoding of PV and FoxP2 neurons in the external globus pallidus during perceptual decision-making

The external globus pallidus (GPe), a central nucleus of the basal ganglia (BG), comprises diverse cell types, including PV and FoxP2 neurons. While their roles in movement regulation and sensory processing have been studied, their contributions to perceptual decision-making remain unclear. Using a Go/No-Go visual task in mice, we found that optogenetic activation of either population impaired behavioral performance, whereas only inhibition of PV neurons, but not FoxP2 neurons, disrupted behavior. Electrophysiological recordings revealed that PV and FoxP2 neurons exhibited distinct encoding of visual stimuli, licking movement, and trial outcomes, with a larger fraction of PV neurons positively encoding the reward-predicting (Go) stimulus and the population overall preferring the rewarded over unrewarded outcome. FoxP2 neuron inhibition had minimal effects on spontaneous firing of neurons in the substantia nigra pars reticulata (SNr), a major BG output. By contrast, PV neuron inhibition altered both spontaneous and task-related activity of SNr neurons, reducing their preference for reward-predicting stimulus or unrewarded outcome in a subpopulation-specific manner. These findings indicate that PV and FoxP2 neurons in the GPe differentially influence sensorimotor processing and SNr activity, with PV neurons playing a key role in supporting accurate performance during perceptual decision-making.

neuroscience↗

bioRxiv2025

MoCETSE: A mixture-of-convolutional experts and transformer-based model for predicting Gram-negative bacterial secreted effectors

Identifying effector proteins of Gram-negative bacterial secretion systems is crucial for understanding their pathogenic mechanisms and guiding antimicrobial strategies. However, existing studies often directly rely on the outputs of protein language models for learning, which may lead to difficulties in accurately recognizing complex sequence features and long-range dependencies, thereby affecting prediction performance. In this study, we propose a deep learning model named MoCETSE to predict Gram-negative bacterial effector proteins. Specifically, MoCETSE first uses the pre-trained protein language model ESM-1b to transform raw amino acid sequences into context-aware vector representations. Then, by employing a target preprocessing network based on a mixture of convolutional experts, multiple sets of convolutional kernel "experts" process the data in parallel to separately learn local motifs and short-range dependencies as well as broader contextual information, generating more expressive sequence representations. In the transformer module, MoCETSE incorporates relative positional encoding to explicitly model the relative distances between residues, enabling the attention mechanism to precisely recognize the sequential relationships and long-range functional dependencies among amino acids, thereby achieving high-accuracy prediction of secreted effectors. MoCETSE has demonstrated outstanding predictive ability in 5-fold cross-validation and independent testing. Benchmark test shows that the performance of MoCETSE surpasses existing excellent binary and multi-class classifiers. Author SummaryGram-negative bacteria inject effector proteins into host cells via secretion systems, disrupting normal cellular functions and inducing diseases. Accurately identifying these virulent proteins is key to understanding bacterial pathogenic mechanisms and developing therapies. However, existing methods face issues like feature redundancy, inadequate capture of long-range dependent signals, and low computational efficiency. We developed MoCETSE, a novel computational method enabling end-to-end intelligent prediction of effector proteins from raw sequences. Due to the high computational cost of position-specific scoring matrix encoding, we use pre-trained protein language models to extract structural, evolutionary, and functional features from sequences, providing biologically meaningful inputs for subsequent deep learning models. Our hybrid convolutional expert network reduces dimensionality of high-dimensional embeddings and extracts multi-scale features, effectively overcoming feature redundancy and information loss, and improving model performance and efficiency. In learning secretion signal features, relative positional encoding models amino acid order, capturing critical long-range dependent signals, and enhancing the biological interpretability of predictions. MoCETSE outperforms existing tools like DeepSecE in cross-category predictions, offering a high-throughput method for effector protein prediction and clues for studying bacterial infections and developing therapies.

bioinformatics↗

bioRxiv2025

Joint decomposition of Hi-C maps reveals salient features of genome architecture across tissues and development

The spatial organization of chromosomes in the nucleus is fundamental to cellular processes. Contact frequency maps from Hi-C and related chromosome conformation capture assays are increasingly available for a wide variety of biosamples and conditions, creating opportunities for comprehensive studies of genome compartmentalization and long-range interactions. However, the conventional dimensionality reduction approach to study long-range contact frequency profiles projects individual datasets into different and incomparable linear subspaces, making the resulting embeddings unsuitable for large-scale integrative analysis. To address this shortcoming and overcome the computational constraints involved in doing so, we introduce an analytic framework and Python toolkit that leverages incremental principal component analysis to project interchromosomal contact frequency profiles across arbitrarily many Hi-C datasets onto a common set of components or basis vectors. Our approach produces robust and directly comparable first and higher-order principal component (PC) scores that collectively capture biologically meaningful information beyond traditional A/B compartments. By applying our framework to a collection of 89 human Hi-C samples, we uncover distinct patterns of nuclear architecture reflecting cell state categories, associated with different heterochromatin state compositions. We also demonstrate that jointly-derived higher-order PCs improve the prediction of gene expression and regulatory element activity during differentiation. Together, our joint decomposition approach provides a powerful and scalable foundation for systematically investigating genome organization, providing critical insights into its role in development and disease.

genomics↗

bioRxiv2025

A synaptic mechanism for encoding the learned value of action-derived safety

Motivated behavior is often framed in terms of biologically grounded outcomes, such as food or threat. Yet many motivated actions, like the pursuit of safety or agency, depend on outcomes that lack explicit sensory value and must instead be inferred from experience. Here, we identify a thalamostriatal circuit mechanism by which such internally constructed outcomes acquire motivational value. In mice performing an active avoidance task, neurons in the paraventricular thalamus (PVT) projecting to the nucleus accumbens (NAc) develop a safety-encoding signal that emerges following successful avoidance. This signal is experience-dependent and value-sensitive, diminishing upon devaluation of the instrumental contingency. Selective silencing of the PVT[->]NAc projection at safety onset disrupts avoidance persistence without impairing action-outcome learning, as confirmed by computational modeling of value updating based on prediction error. Mechanistically, PVT input recruits cholinergic interneurons (CINs) to modulate dopamine release and this influence depends on synaptic potentiation mediated by GluA2-lacking AMPA receptor insertion at PVT-CIN synapses. Disrupting this plasticity selectively impairs the avoidance response by blunting the motivational value of safety without affecting acquisition. These findings reveal how thalamic circuits assign value to abstract, internal outcomes, providing a framework for understanding how goals like safety are inferred, stabilized, and rendered behaviorally effective.

neuroscience↗

bioRxiv2025

Alteration of Water Exchange Rates Following Focused Ultrasound-Mediated BBB Opening in the Dorsal Striatum of Non-Human Primates: A Diffusion-Prepared pCASL Study

This study applied diffusion-prepared pseudo-continuous arterial spin labeling (DP-pCASL) to quantify cerebral blood flow (CBF), arterial transit time (ATT), and blood-brain barrier (BBB) water exchange rate (Kw) before and after focused ultrasound (FUS)-mediated blood-brain barrier opening (BBBO) in the dorsal striatum of four non-human primates. Six baseline and seven BBBO sessions were performed. DP-pCASL was acquired approximately 45 minutes after FUS sonication combined with intravenous microbubbles, and contrast-enhanced T1-weighted imaging was subsequently used to confirm the BBBO region. Whole-brain analyses revealed no significant changes in CBF or ATT following BBBO (permutation p > 0.05). Region-of-interest analysis within the sonicated caudate demonstrated a significant localized decrease in Kw, with median (IQR) values of 45.0 (40.6 - 55.6) min-1 at the BBBO site versus 61.6 (58.3 - 70.4) min-1 in the contralateral control region (p < 0.05), confirming spatially specific suppression of transendothelial water flux. In contrast, whole-brain Kw increased significantly following BBBO, with median (IQR) values of 49.8 (46.3 - 55.9) min-1 in non-BBBO sessions versus 59.4 (56.6 - 66.3) min-1 in BBBO sessions (p < 0.01), indicating a diffuse enhancement of water exchange across the brain. These findings establish DP-pCASL-derived Kw as a sensitive, non-contrast biomarker for both local and global BBB permeability changes induced by focused ultrasound, supporting its potential for longitudinal monitoring in preclinical and clinical neurotherapeutic applications.

neuroscience↗

bioRxiv2025

MViewEMA: Efficient Global Accuracy Estimation for Protein Complex Structural Models Using Multi-View Representation Learning

Estimation of model accuracy (EMA) is crucial for the reliable application of protein structure models in biological research. With the rapid advancement of protein structure prediction techniques and the explosive growth of predicted structural data, existing EMA methods struggle to balance computational efficiency with estimation performance. Here, we present MViewEMA, a single-model EMA method that leverages a multi-view representation learning framework to integrate residue-residue interaction features from micro-environment, meso-environment, and macro-environment levels for global accuracy assessment of protein complex models. Benchmark results demonstrate that MViewEMA outperforms state-of-the-art EMA methods in global accuracy assessment, achieving more than a 10-fold improvement in computational efficiency compared to our previous method, DeepUMQA3. This method enables efficient selection of high-quality protein complex models from large-scale structural datasets and achieved top performance in model selection tracks during the CASP16 blind test, demonstrating its potential to enhance the accuracy of complex structure prediction when integrated into modern frameworks such as AlphaFold-Multimer, AlphaFold3, and DiffDock-PP.

bioinformatics↗

bioRxiv2025

Engineering light robustness: Adaptive evolution uncovers new genetic determinants of HL tolerance in Synechocystis

Excess light absorption causes severe photo-oxidative damage and limits photosynthetic efficiency. Enhancing high-light (HL) tolerance and energy utilization in photosynthetic cyanobacteria holds great potential for sustainable biomanufacturing. Here, using Synechocystis sp. PCC 6803 (Syn6803) as a model, we generated eight independently evolved strains tolerant to hyper-HL (2000 mol photons/m2/s) through approximately two years of natural adaptive laboratory evolution (ALE). Remarkably, these strains circumvented common evolutionary trade-offs; all demonstrated markedly enhanced quantum yields and photosynthetic performance, with four strains exhibiting a 121.71%-168.36% increase in biomass accumulation under HL. When equipped with the heterologous sucrose transporter CscB, sucrose productivity under HL increased by up to 208.33%. Whole-genome resequencing unveiled a total of 77 mutations across the eight lineages, uncovering a diverse polygenic landscape for HL adaptation. Reverse genetics combined with structural modeling highlighted pivotal roles for regulatory and structural loci, including slr0758 (KaiC1) and slr1513 (SbtB), in modulating circadian rhythmicity and inorganic carbon uptake to prevent photo-oxidative overload. Overall, our findings uncover diverse evolutionary routes toward high-light acclimation and provide superior cyanobacterial chassis for photosynthetic biotechnology.

microbiology↗

bioRxiv2025

HSD17B7 is required for Auditory Function by Regulating Cholesterol Synthesis in Sensory Hair Cells

Cholesterol homeostasis is fundamental to cellular function, and its disruption underlies a wide range of human diseases. However, the contribution of cholesterol biosynthesis to auditory physiology remains poorly understood. HSD17B7 (17{beta}-Hydroxysteroid dehydrogenase type 7) catalyzes the conversion of zymosterone to zymosterol, a key step in the post-lanosterol cholesterol biosynthetic pathway. Here, we found that Hsd17b7 is highly enriched in sensory hair cells of zebrafish and mice. The deficiency of Hsd17b7 deficiency reduced intracellular cholesterol levels in HEI-OC1 cells and zebrafish hair cells, thereby compromising MET and acoustic startle responses. A heterozygous nonsense variant (c.544G>T; p.E182*) in HSD17B7 was identified in an individual with bilateral profound hearing loss. mRNA of c.544G>T HSD17B7 failed to rescue the impaired MET and acoustic startle response of hsd17b7 mutants. Mechanistically, the mutation decreases mRNA abundance and significantly reduces protein. Moreover, expression of the p.E182* mutation disrupted the interaction between HSD17B7 and the ER retention receptor RER1, leading to aberrant subcellular localization and altered cholesterol distribution, thereby exacerbating HC dysfunction. Together, our findings suggest a conserved and essential role for HSD17B7-mediated cholesterol biosynthesis in sensory hair cell function and identify HSD17B7 as a candidate gene for sensorineural hearing loss.

developmental biology↗

bioRxiv2025

f-BGM enables fungi-specific genome mining in high accuracy and interpretability

Emerging artificial intelligence (AI)-based genome mining methods have revolutionized the paradigm of bacterial secondary metabolite (SM) discovery. Fortunately, recent data accumulation of fungal biosynthetic gene clusters (BGC) fairly offers opportunities for systematic development and evaluation of fungi-specific pipelines. In this work, we proposed a deep learning framework termed as f-BGM specifically for fungal genome mining. By designing a novel self-attention-based architecture to augment inter-domain associations in local genomic contexts, f-BGM exhibits superior performance over existing AI-based methods in both in-distribution and out-of-distribution benchmark tests for BGC detection. Further analyses demonstrate that f-BGM is of decent interpretability on deciphering single-domain and -protein importance, as well as inter-domain partnership. By establishing additional binary classification models, f-BGM also achieves high-quality identification of core enzymes within given BGCs. Finally, case studies of f-BGM-driven genome mining in marine fungi uncovers biosynthetic potential underestimated by the rule-based method antiSMASH, as supported by experimental and computational validation.

bioinformatics↗

bioRxiv2025

Accurate Protein-Protein Interactions Modeling through Physics-informed Geometric Invariant Learning

AlphaFold has set a new standard for predicting protein structures from primary sequences; however, it faces challenges with protein complexes across species, engineered proteins, and antigen-antibody, where co-evolutionary signals may be sparse or missing. Herein, we present ProTact, a SE(3)-invariant geometric graph neural network that integrates physics-informed geometric complementarity and trigonometric constraints as robust inductive biases to enhance protein-protein contact predictions. ProTact is applicable to both experimental and predicted monomer structures and utilizes a modulated key point matching algorithm to approximate accurate docking poses. Experimental evaluations demonstrate that ProTact consistently outperforms state-of-the-art sequence-based and structure-based methods on benchmark datasets, achieving notable relative improvements of 31.63% in Precision@10 for CASP 13 and 14 targets, and 31.94% for DIPS-Plus datasets. Moreover, when combined with AlphaFold3 as re-scoring functions, ProTact surpasses its default confidence scores, offering over 30.48% improvements in the low-MSA contexts. We anticipate that the proposed framework will advance our understanding of protein interactions, functions, and design.

bioinformatics↗

bioRxiv2025

The Fengshu Large Model for Wugu Fengdeng: An Innovation Engine for Knowledge Integration in the Soybean Field

Against the backdrop of global population growth and the continuous escalation of food demand, the acceleration of agricultural modernization has emerged as the core pathway to safeguard food security. As the worlds fourth-largest food crop, soybean (Glycine max) possesses multiple strategic values for food security, livestock feed, and industrial raw materials, thanks to its high protein content (accounting for over 40% of the dry weight of seeds) and oil resource attributes. However, the soybean industry is confronted with multiple challenges: the long cycle of genetic breeding (8-10 years required by traditional methods), annual losses from diseases and pests reaching 40% (data from FAO 2023), and the reliance on empirical decision-making in field management, all of which urgently call for intelligent solutions.At present, agricultural knowledge is experiencing explosive growth - more than 4,000 new soybean-related literatures are added annually in PubMed, and agricultural technology Q&A platforms (Zhihu, Baidu Tieba, etc.) generate over 3,000 daily questions. However, this multi-source heterogeneous knowledge is scattered in books, literatures, Q&A communities, and gene databases, lacking systematic integration, resulting in a knowledge utilization rate of less than 30%. In response to this, this paper proposes the "Fengshu-Agri" large model, an agricultural knowledge integration and innovation engine based on the collaboration of Retrieval-Augmented Generation (RAG) and Knowledge Graph (KG). The model achieves breakthroughs through the following core technological innovations: O_LIDeep integration of multi-source heterogeneous data: For the first time, a systematic integration of 256 agricultural monographs (covering standardized knowledge such as planting techniques and pest control), 66,772 cutting-edge research literatures (genomics, agronomic trait analysis, etc.), 880,000 production practice Q&A (covering scenarios such as pest diagnosis and environmental stress response), and 120,000 gene annotation data (functional genes, expression regulation mechanisms) has been carried out to construct the largest knowledge graph in the agricultural field - containing 2 million entity nodes, realizing full-chain knowledge modeling from molecular-level gene regulation to field management. C_LIO_LIDouble-layer collaborative retrieval mechanism: Innovatively integrating RAG semantic vector retrieval with KG structured reasoning to solve problems such as low recall rate (<65%) and semantic ambiguity in traditional systems. Specifically, it achieves dual matching through local keywords (entity-level) and global keywords (relationship-level); C_LIO_LIOptimization of domain-specific generation engine: Based on the QWEN-QWQ32B large language model, a three-stage fine-tuning is carried out (pre-training, agricultural instruction fine-tuning, and reinforcement learning from human feedback) [15], which improves the professionalism of generated texts by 35% (manual evaluation index) and significantly reduces the LLM hallucination problem [12]. Experimental results show that Fengshu-Agri achieves an accuracy rate of 89.6% in soybean knowledge retrieval tasks (a 25.4% improvement over traditional RAG models), a recall rate of 87.3% (20.5% higher than pure KG reasoning models), and an F1 score of 88.4%. It can efficiently answer complex questions involving multi-entity associations (such as "the impact mechanism of CRISPR-editing the GmSWEET gene on lepidopteran pest resistance"). Analysis of typical cases indicates that the models responses not only cover technical principles (sgRNA design, Agrobacterium transformation procedures) but also integrate metabolomics data (70% improvement in insect resistance with no significant yield reduction), demonstrating cross-modal knowledge fusion capabilities. In the future, this model will be expanded to crops such as maize and potato to construct an agricultural intelligent ecosystem covering the entire process of cultivation management and breeding improvement, promoting the upgrading of precision agriculture toward a data-driven model. C_LI

plant biology↗

bioRxiv2025

Identifying Optimal Machine Learning Approaches for Microbiome-Metabolomics Integration with Stable Feature Selection

Microbiome research is often limited by methodological inconsistencies that reduce reproducibility and functional insight. Traditional taxonomy-based profiling is limited by sparse data, variable resolution, and reliance on gDNA sequencing which provides only indirect links to microbial function. Multi-omics integration offers a framework for linking community composition to functional outputs, but progress has been hindered by the lack of standardized frameworks and inconsistent use of machine learning. In particular, feature selection stability, which is central to biomarker discovery and experimental validation, remains underexplored. Here, we systematically benchmarked three widely used algorithms (Elastic Net, Random Forest, XGBoost) across seven multi-omics integration strategies and single-omics models. Additionally, we evaluated the impact of transforming metabolomics and taxonomic abundance data. Using human gut microbiome datasets that integrate metagenomic taxonomic profiles with metabolomics, we evaluated models for 9 binary and 8 continuous outcomes across 20 train/test splits per dataset. We further assessed the effect of feature reduction on both predictive accuracy and feature selection stability. Nonlinear learners were most consistently competitive: continuous outcomes favored metabolomics-dominant models, whereas binary outcomes favored stacked multi-omics models. Random Forest and XGBoost also yielded greater feature selection stability, particularly for full dimensional metabolomics data. Together, these findings demonstrate how integration strategy, algorithm choice, and data preprocessing jointly shape predictive performance and feature selection reproducibility in multi-omics microbiome modeling.

bioinformatics↗

bioRxiv2025

Spatial determinants of tumor cell dedifferentiation and plasticity in primary cutaneous melanoma

Localized cutaneous melanoma can be cured by excision but success critically depends on early detection and risk assessment of primary lesions. However, their initiation, progression, and immunology remain poorly understood, partly due to high intra- and inter-tumor heterogeneity. We studied this heterogeneity using spatial profiling in over 300 histological domains, each representing a single progression stage, and found that 200-600 cell neighborhoods from a single melanoma can be as different in RNA and protein expression as neighborhoods from different tumors. These differences are not stochastic, however, and disease progression can be mapped at the neighborhood level onto a cell state landscape defined by the activity of the melanocyte master regulator MITF and genes associated with a dedifferentiated neural crest phenotype. Position in this landscape is influenced by proximity to immune cells, perivascular environments, and other tissue features, but no single association is absolute, giving rise to complex spatial patterns.

cancer biology↗

bioRxiv2025

Investigating diurnal effects and joint nest defense behavior in Herring Gulls

Nest defense in birds is vital for the protection of their young, but can prove energetically costly. Birds often show plasticity in nest defense depending on factors such as threat level, mate presence, nest stage, and body condition. Such factors can vary over temporal extents ranging from a diel cycle to a breeding season or an individuals lifetime. Understanding diurnal variation in nest defense intensity can be particularly useful when studying a breeding population to help investigators minimize nest disturbance, yet few studies have explored diurnal variation in nest defense intensity. Here we investigated how nest defense by Herring Gulls (Larus smithsonianus) varies with time of day. We simulated predatory threat at 26 Herring Gull nests during four different times of day: "Early Morning" (0530-0600), "Late Morning" (1000-1030), "Afternoon" (1400-1430), and "Evening" (1900-1930). Based on previous findings of diurnal activity patterns in a Herring Gull, we predicted that nest defense intensity--measured as aggressiveness of response and latency to calm--would be greater during the early morning and late evening than at other times. Contrary to our prediction, we found that the time of day did not affect nest defense intensity in Herring Gulls. However, independent of time of day, we found that when both mates were present at the nest, aggression scores were elevated. Our results suggest that joint nest defense in Herring Gulls permits greater aggressiveness towards predators, perhaps due to division of labor or lowered risk of complete nest failure if one parent is injured or killed. Further, our results indicate that researchers can minimize nest disturbance and accompanying stress by limiting research activities at Herring Gull nests when both parents are present.

animal behavior and cognition↗

bioRxiv2025

Deciphering the microbial landscape of lower respiratory tract in health and respiratory diseases

The microbiota plays an important role in maintaining lung homeostasis and the development of respiratory diseases. However, there have not been detailed characteristics of lung microbiome in health and respiratory diseases. Bronchoalveolar lavage fluid of 43 healthy individuals, 23 COVID patients, and 28 non-small cell lung cancer patients was conducted to characterize microbial diversity and function by metagenomic sequencing. We identified 196 species in health lung after removing low abnudance species. The most abundant species were Staphylococcus aureus, Salmonella enterica, and Klebsiella pneumoniae. Keystone species were identified through co-network analysis, such as Granulicatella adiacens and Mogibacterium diversum. To further explore the microbial function in the healthy lung, we obtained 56 non-redundant metagenome-assembled genomes (MAGs) and subsequently classified them into nine phyla. Metabolism and genetic information processing exhibit high abundance, including bacterial secretion systems, protein export, and Staphylococcus aureus infection, indicating a close correlation between microbiota and the host. Furthermore, the microbiome pairwise distances of non-small cell lung cancer (NSCLC) participants were more similar to those of the healthy group than those of individuals with COVID-19. Rothia dentocariosa, Corynebacterium kefirresidentii, Sphingobium yanoikuyae were enriched in the healthy group, while Schaalia odontolytica, Actinomyces graevenitzii, and Rothia mucilaginosa was enriched in the NSCLC group. Overall, this study provides a comprehensive overview of the diversity and gene function of the lung microbiome, which will facilitate future studies of microbiota associated with respiratory diseases in humans.

microbiology↗