Biology subjects

Pollard, K. S.

Publications and source records attributed to Pollard, K. S..

At least 19 recordsLinked to original sources

bioRxiv2026

ChAI-Net learns a regulatory code for chromatin co-accessibility

The regulation of chromatin accessibility is critical for cell function and identity. Advances in single-molecule long-read footprinting have enabled near-nucleotide-resolution readout of chromatin fiber organization, capturing both coordination of transcription factor (TF) binding and variation in nucleosome positioning previously obscured by bulk methods. Here we combine this technique with deep learning to study how protein-DNA interactions are coordinated within different types of cis-regulatory elements in mouse embryonic stem cells. First, we quantify the frequency with which pairs of positions are simultaneously accessible on individual molecules. Then, we train ChAI-Net, a neural network that predicts these locus-specific chromatin co-accessibility maps from the underlying DNA sequence. Through model interpretation and in silico experiments, we show that ChAI-Net learns unique combinations of sequence motifs underlying co-accessibility within enhancers, promoters, and CTCF-bound sites. Many of these motifs are missed when modeling average accessibility, suggesting a specific role in coordinated binding. ChAI-Net's regulatory grammar also highlights the importance of TF motif orientation, spacing, positioning, and co-occurrence in driving co-accessibility. Thus, our framework for extracting regulatory rules from footprinting data expands understanding of TF cooperativity, while laying the groundwork to elucidate how these rules vary across cellular contexts as single-molecule genomics is deployed at scale.

genomics↗

bioRxiv2026

Machine learning reveals sequence and genomic context features underlying Alu-specific effects on genome folding

The Alu transposable element is among the most abundant classes of mobile DNA in the human genome, and has been linked to gene regulation and chromatin organization. Yet how individual Alu insertions influence nearby chromatin interactions remains poorly understood. To investigate this, we used deep learning to perform a genome-wide in silico deletion screen of ~1.1 million Alus, predicting each element's importance to local genome folding. We identified a subset of high-scoring Alus that span multiple Alu subfamilies and are enriched in loci that are fast-evolving and gene-dense, especially loci encoding genes that are actively transcribed and/or related to Alu biology. We further found that polymorphic Alus preferentially occur in regions tolerant of sequence variation but predicted to be resistant to changes in chromatin structure. Targeted in silico mutagenesis showed that the importance of individual Alus to local chromatin interactions depends on both intrinsic Alu sequence properties and genomic context. Finally, we identified Alu sequences predicted to alter CTCF-mediated boundary strength and, in some cases, to promote the formation of new loops and boundaries. Together, these results position Alus as key modulators of genome architecture, while underscoring that the fate of a new Alu depends on where it inserts and how it interacts with other determinants of chromatin state and structure.

genomics↗

bioRxiv2026

Causal variant underestimation is a major overlooked driver of sequence-to-function model underperformance

Deep learning sequence-to-function (S2F) models represent tremendous promise for functionally fine-mapping causal variants associated with traits and disease. Yet it remains unclear precisely how effective they are at this task. Generally, S2F models perform well at classifying putatively causal expression quantitative trait loci (eQTL) SNVs, yet dramatically underperform linear baselines at ranking different individuals' gene expression values from their whole genome sequence, which directly calls into question their ability to fine-map a locus by properly weighting the effects of variants onto gene expression. Here, using the state-of-the-art S2F model AlphaGenome, we systematically compared predicted effect sizes for fine-mapped eQTLs to nearby putatively non-causal SNVs, finding that AlphaGenome pervasively underestimates the effects of most causal variants and fails to successfully fine-map eQTLs in most loci for this reason. We show that misdirected variant effect predictions and negative cross-individual correlations, widely cited as major challenges facing S2F expression modeling, are not egregious errors but a chance consequence of weak, noisy attributions when causal variants are underestimated. Our results suggest that a failure to detect local variant effects onto enhancer activity is a cause of underestimation. Additionally, we find that underestimated variants are enriched far from their target gene, suggesting inadequate enhancer-gene linking causes underestimation even when local effects are well-detected. Finally, our results clarify when S2F models are reliable for fine-mapping objectives: while they pervasively underestimate most causal variants, demonstrating limited fine-mapping utility for most loci, the top 0.1 most prominent variant effect predictions are strongly enriched for causal variants with directionally correct predictions that are consistent across model replicates, suggesting extremely strong predictions are broadly trustworthy. Our results demonstrate that causal variant underestimation is a core issue facing S2F expression predictors, with future improvements dependent on better local activity detection and enhancer-gene linking.

genetics↗

bioRxiv2026

Quantitative detection of gut microbial eukaryotes with EukDetect2 reveals global distribution of commensal protists and association with distinct microbial community structure

Microbial eukaryotes are prevalent members of host-associated and free-living microbial communities, but are routinely excluded from studies of these communities. Existing methods for eukaryote detection from whole metagenome sequencing are limited by contamination of eukaryotic reference genomes and incomplete taxonomic coverage. Our previously published tool EukDetect addressed these challenges using a curated database of universal BUSCO marker genes, but lacked validated quantitative abundance metrics and was built from a limited number of genomes. Here we present EukDetect2, incorporating a database containing 6,948 microbial eukaryotic genomes representing 6,594 unique species, 2,339 of which are newly added since EukDetect version 1, alongside quantitative metrics for estimating absolute and relative abundance of microbial eukaryotes. Using simulated data, we demonstrate accurate abundance estimation, no false positives from bacterial or host-derived reads, and equivalent or greater sensitivity and specificity than alternative taxonomic profiling tools across a range of microbial abundances and community compositions. Applying EukDetect2 across globally distributed human gut microbiome cohorts, we find that Blastocystis spp. and Dientamoeba fragilis are the most prevalent gut eukaryotes across cohorts, while host-associated fungi are consistently less prevalent than commensal protists. Blastocystis abundance is positively associated with a gut microbial community enriched for fiber-fermenting microbes and depleted for pro-inflammatory and industrialization-associated taxa. EukDetect2 provides sensitive, accurate, and quantitative metrics for investigating microbial eukaryotes from metagenomic samples.

microbiology↗

bioRxiv2026

Gut microbiome-dependent IL-1 signaling is a mediator of ACVR1R206H-driven heterotopic ossification

Inflammatory diseases cause significant morbidity and mortality, but their pathobiology is often difficult to dissect due to complex genetic-environmental interactions. Genetic forms of heterotopic ossification, such as fibrodysplasia ossificans progressiva (FOP), reduce genetic variability, allowing careful dissection of non-genetic drivers of inflammation. While >95% of FOP patients harbor the ACVR1R206H mutation, patients exhibit significant variability in disease progression, suggesting a role of environmental drivers. Here, we identify the gut microbiome as a regulator of inflammation-driven HO in FOP. Metagenomic profiling of cohabitating FOP/unaffected sibling pairs revealed a pathogenic gut microbiome profile in FOP patients (Bray-Curtis, p < 0.05). In Pdgfr-Cre/Acvr1R206H (FOP) mice, gut microbiome ablation by antibiotics reduced spontaneous HO formation (47.4% reduction, p < 0.05) and reduced plasma IL-1 pathway activity. IL-1{beta} blockade in FOP mice suppressed trauma-induced HO formation. These findings identify a gut microbiome-IL-1-HO axis with modifiable targets for developing treatments for HO and related inflammatory conditions. One Sentence SummaryAntibiotic disruption of the gut microbiome reduces HO in FOP mice via an IL-1 mediated pathway.

immunology↗

bioRxiv2026

Structural variants in human congenital heart disease disrupt distal genomic regulatory contacts of developmental genes

Predicting the functional significance of structural variants (SVs) associated with genetic diseases remains challenging. To test the hypothesis that SVs from people with congenital heart disease (CHD) disrupt developmental chromatin interactions, we developed CardioAkita, a machine-learning model that predicts how variants alter 3D chromatin structure. Analyzing previously genotyped de novo SVs (dnSVs), we observed a positive association between CHD severity and CardioAkita scores across dozens of families. From whole-genome sequencing of three individuals with CHD we predicted disruptive dnSVs. Induced pluripotent stem cells engineered to harbor these variants confirmed CardioAkitas predictions of 3D chromatin changes, and further revealed aberrant expression of local genes including cardiac developmental genes, suggesting that chromatin reorganization plays a significant mechanistic role in the genetic etiology of CHD. Our findings highlight the potential for models of 3D chromatin organization to predict the pathogenicity and underlying mechanisms of SVs in human disease.

genomics↗

bioRxiv2025

Bacterial strains in the human gut associate with host physiology

The human microbiota influences host physiology, yet much of its complexity lies beneath the species level. Here, we analyzed the intra-species genetic diversity of 936 gut bacteria across 24,997 individuals from three countries. Our findings show that highly abundant species exhibit greater strain stability, whereas low-abundance species display increased in-clonal mutations. Clonal strains are often mutually exclusive, while genetically variable strains tend to coexist. Strain turnover is associated with the presence of annotated chemotaxis and sporulation genes in reference genomes, whereas strain coexistence is associated with quorum sensing and secretion systems. Leveraging deep phenotypic data, we constructed an atlas detailing strain-level associations with diverse host physiological domains. For example, Phocaeicola vulgatus sub-types relate to host obesity, Lachnospira eligens to sleep, and Parabacteroides distasonis to iron hemostasis. This resource may guide personalized microbiome-based interventions to improve human health.

microbiology↗

bioRxiv2025

Self-supervised learning enables robust microbiome predictions in data-limited and cross-cohort settings

The gut microbiome plays a crucial role in human health, but machine learning applications in this field face significant challenges, including limited labeled data availability, high dimensionality, and batch effects across different cohorts. To address these limitations, we developed representation learning models for gut microbiome metagenomic data, drawing inspiration from foundation models approaches based on self-supervised and transfer learning principles. By leveraging a large collection of 85,364 metagenomic samples, we implemented multiple self-supervised learning methods, including masked autoencoders with varying masking rates and adapted single-cell RNA sequencing models (scVI and scGPT), to generate embeddings from bacterial abundance profiles. These learned representations demonstrated significant advantages over raw bacterial abundances in two key scenarios: first, when training predictive models with very limited labeled data, improving prediction performance for age (r = 0.14 vs. 0.06), BMI (r = 0.16 vs. 0.11), visceral fat mass (r = 0.25 vs. 0.18), and drug usage classification (PR-AUC = 0.81 vs. 0.73); and second, when generalizing predictions across different cohorts, consistently outperforming models based on raw abundances in cross-dataset evaluation. Our approach provides a valuable framework for leveraging self-supervised representation learning to overcome the data limitations inherent in microbiome research, potentially enabling more robust and generalizable machine learning applications in this field.

systems biology↗

bioRxiv2025

Expanding vaginal microbiome pangenomes via a custom MIDAS database reveals Lactobacillus crispatus accessory genes associated with cervical dysplasia

The vaginal microbiome plays a central role in reproductive health. Vaginal microbiome dysbiosis is associated with many adverse reproductive health outcomes, but most studies have focused on associations at the species level. The potential contribution of intraspecies microbial variation, especially gene content differences across bacterial strains, remains underexplored in reproductive health contexts. The Metagenomic Intra-Species Diversity Analysis (MIDAS) framework enables such analyses but depends on comprehensive reference databases. We constructed a MIDAS-compatible pangenome database from over 18,000 genomes in the Vaginal Microbiome Genome Collection (VMGC). Compared to the Genome Taxonomy Database (GTDB)-derived reference, the VMGC database expanded pangenomes of prevalent vaginal species, better capturing vaginal-specific intraspecies diversity. Applying this database to vaginal samples from a cervical dysplasia cohort, we identified thirteen Lactobacillus crispatus accessory genes significantly associated with cervical dysplasia, including a HicAB toxin-antitoxin system, three transcriptional regulators, and three phage-derived genes. These findings highlight the utility of body site-specific reference resources and shotgun metagenomic sequencing for uncovering intraspecies microbial variation relevant to reproductive health. IMPORTANCEThe vaginal microbiome plays a critical role in reproductive health, and different bacteria from the same species can carry different genes that influence how the strains interact with the host and other microbes. These strain-level differences are often overlooked when microbiomes are analyzed only at the species level. Existing genomic reference databases are heavily biased toward gut and environmental bacteria, leaving the genetic diversity of vaginal microbes understudied. We built a specialized reference database from over 18,000 vaginal bacterial genomes that better reflects this diversity. We then applied this resource to quantify gene-level variation in vaginal samples from a cervical dysplasia cohort. Focusing on Lactobacillus crispatus, a prevalent and often beneficial vaginal species, we identified thirteen genes that were more common in women with cervical dysplasia than in controls. This work demonstrates that body site-specific genomic resources are essential for uncovering strain-level bacterial differences relevant to reproductive health.

microbiology↗

bioRxiv2025

Comparative evaluation of genomic footprinting algorithms for predicting transcription factor binding sites in single-cell data

Transcription factors (TFs) have millions of potential binding sites across the human genome, but only a fraction are bound in a given context. Genomic footprinting aims to identify context-specific binding sites by detecting patterns in open chromatin data. While powerful, these approaches face technical challenges, especially in single-cell applications. We developed a benchmarking framework for cell-type specific footprinting and used it to evaluate the consistency, reproducibility, and equivalency of three leading methods across data quality scenarios and as a function of cell-type similarity. Peak-level read coverage emerged as the strongest predictor of stable footprints. Motivated by limited reproducibility across tools, we built an ensemble model that improved concordance with ChIP-seq. To encourage broader adoption and continued tool development, we provide practical guidelines for robust genomic footprinting in single-cell datasets and a roadmap for extracting deeper insights about how gene regulatory networks vary across cell types in complex tissues.

genomics↗

bioRxiv2025

GAME: Genomic API for Model Evaluation

The rapid expansion of genomics datasets and the application of machine learning has produced sequence-to-activity genomics models with ever-expanding capabilities. However, benchmarking these models on practical applications has been challenging because individual projects evaluate their models in ad hoc ways, and there is substantial heterogeneity of both model architectures and benchmarking tasks. To address this challenge, we have created GAME, a system for large-scale, community-led standardized model benchmarking on user-defined evaluation tasks. We borrow concepts from the Application Programming Interface (API) paradigm to allow for seamless communication between pre-trained models and benchmarking tasks, ensuring consistent evaluation protocols. Because all models and benchmarks are inherently compatible in this framework, the continual addition of new models and new benchmarks is easy. We also developed a Matcher module powered by a large language model (LLM) to automate ambiguous task alignment between benchmarks and models. Containerization of these modules enhances reproducibility and facilitates the deployment of models and benchmarks across computing platforms. By focusing on predicting underlying biochemical phenomena (e.g. gene expression, open chromatin, DNA binding), we ensure that tasks remain technology-independent. We provide examples of benchmarks and models implementing this framework, and anticipate that the community will contribute their own, leading to an ever-expanding and evolving set of models and evaluation tasks. This resource will accelerate genomics research by illuminating the best models for a given task, motivating novel functional genomic benchmarks, and providing a more nuanced understanding of model abilities.

bioinformatics↗

bioRxiv2025

Linkage of nucleotide and functional diversity varies across gut bacteria.

Understanding the forces shaping genomic diversity within bacterial species is essential for interpreting microbiome evolution, ecology, and host associations. Here, we analyze over one hundred prevalent gut bacterial species using the Unified Human Gut Genome (UHGG) collection to characterize patterns of intra-specific genomic variability. Gene content divergence scales predictably with divergence in core genome single nucleotide polymorphisms (SNPs), though there is substantial variability in evolutionary dynamics across species. Overall, accessory genes exhibit consistently faster linkage decay compared to core SNPs, highlighting the fluidity of functional repertoires within species boundaries. This signal is strongest for mobile genetic elements, which show minimal linkage to core genome SNPs. Together, our findings reveal species-specific recombination regimes in the gut microbiome, underscoring the importance of accounting for horizontal gene transfer and genome plasticity in microbiome-wide association studies and evolutionary models.

microbiology↗

bioRxiv2025

Improving interpretability of transcription factor binding models with DNA shape features

Deep learning models in genomics that predict molecular phenotypes from DNA sequence traditionally focus on one-hot encoded nucleotide representations. Here, we develop a novel model that extends this approach by incorporating DNA structural attributes indicative of local DNA shape alongside canonical sequence inputs. This augmentation provides an additional axis for model interpretability and aids in identifying regulatory patterns not apparent from sequence alone. Applying this approach to prediction of transcription factor binding (ChIP-seq) demonstrates that combining sequence and structural DNA information can improve the identification of regulatory elements to provide a more nuanced understanding of genomic function and regulation. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=96 SRC="FIGDIR/small/646034v1_ufig1.gif" ALT="Figure 1"> View larger version (23K): org.highwire.dtl.DTLVardef@b648b6org.highwire.dtl.DTLVardef@15fa931org.highwire.dtl.DTLVardef@15d0d31org.highwire.dtl.DTLVardef@d212e_HPS_FORMAT_FIGEXP M_FIG C_FIG Schematic overview of the DeepShape model. One-hot encoded sequence, and five DNA shape attributes--minor groove width (MGW), helical twist (HelT), propeller twist (ProT), roll, and electrostatic potential (EP)--are input separately to sequence and shape branches of the model, each processed through two convolutional layers and subsequently concatenated for further processing through additional convolutional layers.

genomics↗

bioRxiv2025

Machine learning-predicted chromatin organization landscape across pediatric tumors

Structural variants (SVs) are increasingly recognized as important contributors to oncogenesis through their effects on 3D genome folding. Recent advances in whole-genome sequencing have enabled large-scale profiling of SVs across diverse tumors, yet experimental characterization of their individual impact on genome folding remains infeasible. Here, we leveraged a convolutional neural network, Akita, to predict disruptions in genome folding caused by somatic SVs identified in 61 tumor types from the Childrens Brain Tumor Network dataset. Our analysis reveals significant variability in SV-induced disruptions across tumor types, with the most disruptive SVs coming from lymphomas and sarcomas, metastatic tumors, and germline cell tumors. Dimensionality reduction of disruption scores identified five recurrently disrupted regions enriched for high-impact SVs across multiple tumors. Some of these regions are highly disrupted despite not being highly mutated, and harbor tumor-associated genes and transcriptional regulators. To further interpret the functional relevance of high-scoring SVs, we integrated epigenetic data and developed a modified Activity-by-Contact scoring approach to prioritize SVs with disrupted genome contacts at active enhancers. This method highlighted highly disruptive SVs near key oncogenes, as well as novel candidate loci potentially implicated in tumorigenesis. These findings highlight the utility of machine learning for identifying novel SVs, loci, and genetic mechanisms contributing to pediatric cancers. This framework provides a foundation for future studies linking SV-driven regulatory changes to cancer pathogenesis.

genomics↗

bioRxiv2025

Dose-dependent sensitivity of human 3D chromatin to a heart disease-linked transcription factor

Dosage-sensitive transcription factors (TFs) underlie altered gene regulation in human developmental disorders, and cell-type specific gene regulation is linked to the reorganization of 3D chromatin during cellular differentiation. Here, we show dose-dependent regulation of chromatin organization by the congenital heart disease (CHD)- linked, lineage-restricted TF TBX5 in human cardiomyocyte differentiation. Genome organization, including compartments, topologically associated domains, and chromatin loops, are sensitive to reduced TBX5 dosage in a human model of CHD, with variations in response across individual cells. Regions normally bound by TBX5 are especially sensitive, while co-occupancy with CTCF partially protects TBX5-bound TAD boundaries and loop anchors. These results highlight the importance of lineage-restricted TF dosage in cell-type specific 3D chromatin dynamics, suggesting a new mechanism for TF-dependent disease.

genomics↗

bioRxiv2024

De novo structural variants in autism spectrum disorder disrupt distal regulatory interactions of neuronal genes

Three-dimensional genome organization plays a critical role in gene regulation, and disruptions can lead to developmental disorders by altering the contact between genes and their distal regulatory elements. Structural variants (SVs) can disturb local genome organization, such as the merging of topologically associating domains upon boundary deletion. Testing large numbers of SVs experimentally for their effects on chromatin structure and gene expression is time and cost prohibitive. To address this, we propose a computational approach to predict SV impacts on genome folding, which can help prioritize causal hypotheses for functional testing. We developed a weighted scoring method that measures chromatin contact changes specifically affecting regions of interest, such as regulatory elements or promoters, and implemented it in the SuPreMo-Akita software (Gjoni and Pollard 2024). With this tool, we ranked hundreds of de novo SVs (dnSVs) from autism spectrum disorder (ASD) individuals and their unaffected siblings based on predicted disruptions to nearby neuronal regulatory interactions. This revealed that putative cisregulatory element interactions (CREints) are more disrupted by dnSVs from ASD probands versus unaffected siblings. We prioritized candidate variants that disrupt ASD CREints and validated our top-ranked locus using isogenic excitatory neurons with and without the dnSV, confirming accurate predictions of disrupted chromatin contacts. This study establishes disrupted genome folding as a potential genetic mechanism in ASD and provides a general strategy for prioritizing variants predicted to disrupt regulatory interactions across tissues.

bioinformatics↗

bioRxiv2024

Oxytocin receptor controls distinct components of pair bonding and development in prairie voles

Oxytocin receptor (Oxtr) signaling influences complex social behaviors in diverse species, including social monogamy in prairie voles. How Oxtr regulates specific components of social attachment behaviors and the neural mechanisms mediating them remains unknown. Here, we examine prairie voles lacking Oxtr and demonstrate that pair bonding comprises distinct behavioral modules: the preference for a bonded partner, and the rejection of novel potential mates. Our longitudinal study of social attachment shows that Oxtr sex-specifically influences early interactions between novel partners facilitating the formation of partner preference. Additionally, Oxtr suppresses promiscuity towards novel potential mates following pair bonding, contributing to rejection. Oxtr function regulates coordinated patterns of gene expression in regions implicated in attachment behaviors and regulates the expression of oxytocin in the paraventricular nucleus of the hypothalamus, a principal source of oxytocin. Thus, Oxtr controls genetically separable components of pair bonding behaviors and coordinates development of the neural substrates of attachment.

neuroscience↗

bioRxiv2024

Deep-learning prediction of gene expression from personal genomes

Models that predict RNA levels from DNA sequences show tremendous promise for decoding tissue-specific gene regulatory mechanisms1-5, revealing the genetic architecture of traits6-10, and interpreting noncoding genetic variation10,11. Existing methods take two different approaches: 1) associating expression with linear combinations of common genetic variants (training across individuals on single genes)12,13, or 2) learning genome-wide sequence-to-expression rules with neural networks (training across loci using a reference genome)11,14,15. Since limitations of both strategies have been highlighted recently16-20, we sought to combine the sequence context provided by deep learning with the information provided by cross-individual training. We utilized fine-tuning to develop Performer, a model with accuracy approaching the cis-heritability of most genes. Performer prioritizes genetic variants across the allele frequency spectrum that disrupt motifs, fall in annotated regulatory elements, and have functional evidence for modulating gene expression. While obstacles remain in personalized expression prediction, our findings establish deep learning as a viable strategy.

genetics↗