Biology subjects

Yuan, K.

Publications and source records attributed to Yuan, K..

At least 19 recordsLinked to original sources

bioRxiv2026

NeoOmni Atlas: an integrated database for pan-cancer neoantigen derived from proteogenomics analysis

Tumor neoantigens derived from diverse molecular aberrations are important targets for personalized cancer immunotherapy, but existing resources incompletely integrate mass-spectrometry-supported canonical and non-canonical antigen evidence across cancer types. NeoOmni Atlas is an interactive pan-cancer T-cell epitope resource, built via proteogenomic reanalysis of 7,561 immunopeptidomics datasets comprising 511 million raw spectra from 2,341 samples across 26 cancer types. The analysis yielded 872,414 eluted canonical peptides and 198,961 non-canonical peptides spanning 18 molecular alteration classes, along with 759,386 unique ligands presented by 322 human leukocyte antigen (HLA) alleles. Relative to existing repositories, NeoOmni Atlas covers a broader range of cancer types, includes 60 additional HLA alleles, and catalogs 369,781 novel ligands. A computational tumor-specific filtering framework identified 316,126 tumor-associated epitopes and 61,231 tumor-specific epitopes. Tumor-specific epitopes showed stronger predicted binding affinity than tumor-associated epitopes for HLA class I A and B alleles, and 65 conserved tumor-specific epitopes recurrent across more than 10 tumor types represent potential shared antigen candidates. NeoOmni Atlas also prioritizes candidate tumor-associated antigens including PMEL (gp100:209-217) and CSPG4, and supports browsing, searching, and downloading by cancer type, human leukocyte antigen allele, peptide sequence, and source event. NeoOmni Atlas is freely accessible at http://modinfor.com/NeoOmni-Atlas/.

bioinformatics↗

bioRxiv2026

Self-Supervised AI Discovery of Histomorphological Phenotypes from Routine Mesothelioma Biopsies

Accurate subtype diagnosis is essential for guiding therapy and predicting patient outcome in malignant mesothelioma. Most computational pathology models are trained on large tissue images from resection specimens, which maximises information for training but limits model relevance in real-world diagnostic settings where small biopsies are the most usual tissue source. In this work, we assembled a large multicentre cohort of HES- and HPS-stained mesothelioma biopsy slides. We used a self-supervised learning model to evaluate the associations of biopsy-driven morphology patterns with histological subtype, molecular markers, and survival. The discovered histomorphology patterns captured a continuum of tissue phenotypes spanning epithelioid, sarcomatoid, and non-tumour morphologies. Also, patient-level HPC representations achieved excellent performance for distinguishing epithelioid from non-epithelioid mesothelioma (AUC = 0.94) and demonstrated predictive value for immunohistochemistry (IHC) markers. Additionally, HPC-derived features alone achieved performance comparable to established clinical and molecular variables (C-index = 0.65), while integration of HPCs with clinical and marker information improved performance to a C-index of 0.69. Several HPCs were significantly associated with favourable or adverse prognosis and reflected known subtype-specific biological patterns. In conclusion, self-supervised learning can discover interpretable histomorphological phenotypes directly from routine mesothelioma biopsies without further training. These AI-derived phenotypes capture clinically and biologically relevant information, linking tissue architecture to molecular characteristics, histological subtypes, and patient outcomes. The proposed framework provides a thorough evaluation of real-world biopsy data using a pre-trained model, without the need for computationally intensive retraining, and addresses the question of whether SSL-based AI can be deployed out of the box in clinical settings.

cancer biology↗

bioRxiv2026

Genome-wide association study of grain iron and zinc concentrations in a diverse CIMMYT wheat panel across contrasting moisture environments

Micronutrient deficiencies remain a major public-health challenge, and genetic improvement of grain iron and zinc concentrations in wheat offers a sustainable biofortification strategy. We evaluated 563 CIMMYT advanced wheat lines and six checks under restricted irrigation (two irrigations) and well-watered conditions (five irrigations) at Ciudad Obregon, Mexico. Grain iron and zinc concentrations were quantified by energy-dispersive X-ray fluorescence spectrometry. Best linear unbiased estimates were calculated, and genome-wide association analyses were conducted using 8,687 high-quality single-nucleotide polymorphisms and a multi-locus mixed model that accounted for population structure. Five marker-trait associations were detected in at least two datasets. For grain zinc concentration, S3B_811421507 was detected under both irrigation regimes and in the combined analysis, whereas S3B_814372642 was detected under well-watered conditions and in the combined analysis. For grain iron concentration, S2B_73321328 and S4B_20679079 were associated with variation under restricted irrigation and in the combined analysis, while S2B_72162723 was detected under well-watered conditions and in the combined analysis. Individual loci explained 0.34% to 3.36% of phenotypic variance, consistent with the quantitative inheritance of grain micronutrient concentration. The identified alleles provide candidate targets for validation and marker-assisted biofortification breeding, while their environment-dependent effects emphasize the need to evaluate micronutrient traits across contrasting moisture conditions.

plant biology↗

bioRxiv2026

Image-based Disease-wide Association Study via Self-supervised Learning links Abdominal MRI Features to 158 Diseases

Medical images contain rich phenotypic information that is often not fully captured by manual clinical assessment. Here, we present a systematic framework for extracting such information and conducting image-based disease-wide association studies (iDWAS) using self-supervised learning (SSL). We applied this framework to abdominal MRI data from the UK Biobank and evaluated associations between MRI-derived features and 562 diseases. Features were learned using a VICReg-based SSL model and tested for disease associations using logistic regression models. We identified 158 diseases that were significantly associated with the MRI-derived features, including the ones which are not directly linked to abdominal anatomy. Focusing on Metabolic dysfunction-associated steatohepatitis (MASH), we showed that the MRI-derived features captured disease-relevant information and separated MASH cases from controls better than established biomarkers such as PDFF and Iron-cT1. These findings highlight the ability of SSL to uncover clinically meaningful signals from routine imaging data. The proposed framework is broadly applicable to other imaging datasets and modalities, enabling more systematic approaches to incidental and early disease detection. The trained model and analysis pipeline are publicly available at https://github.com/srm2022/iDWAS. BackgroundMedical images, often analyzed manually by clinicians, contain valuable information that may not be fully captured by the human eye. Machine Learning (ML) has demonstrated significant potential in extracting this information, allowing us to enhance our understanding of underlying pathologies. We present a systematic framework for extracting such information and performing image-based disease-wide association studies (iDWAS) using self-supervised learning. We then applied the framework to abdominal Magnetic Resonance Imaging (MRI) from UK Biobank (UKB) and investigated the associations between MRI-derived features and 562 diseases. Finally, we delved deeper into the observed associations for Metabolic dysfunction-associated steatohepatitis (MASH), a prevalent yet underdiagnosed condition with a high unmet need for non-invasive diagnostic tools. MethodsWe extracted features from MRI images using VICReg-based self-supervised learning and modeled the associations of the extracted features with 562 diseases using logistic regression. Using the likelihood ratio (LR) test, we then assessed how much additional variance can be explained by the MRI-derived features relative to the variance explained by a null model including only common confounders such as age, gender, and BMI. ResultsOf these 562 diseases, we identified 158 diseases to be significantly associated with the MRI-derived features. While many of them (e.g. MASH) are known to primarily emerge in the abdomen, others (e.g. neuropsychiatric diseases) are not. Taking MASH as a use-case and adjusting for MASH-specific confounding, we then showed that our MRI-derived features captured additional MASH-relevant information compared to standard MRI-derived biomarkers of liver health such as proton density fat fraction (PDFF) and iron-corrected T1 (Iron-cT1). The trained SSL model and analysis pipeline are available at https://github.com/srm2022/iDWAS. ConclusionsOur results illustrate (1) how MRIs from one organ of the body can inform about the diseases primarily rooted in other organs of the body, reflecting patients overall health status, and (2) how SSL can extract more information relevant for specific diseases compared to current state-of-the-art biomarkers. Our framework can be applied to other MRI datasets as well as other imaging modalities with minimal adjustment, providing opportunities for the development of more systematic approaches to incidental diagnosing, i.e. (early) detection of diseases falling outside the scope of the original imaging procedure.

bioinformatics↗

bioRxiv2026

Protein language models learn underlying mutation biases alongside fitness landscapes

Protein language models (PLMs) score the effects of amino acid replacements as pseudo-probabilities, which are widely utilised to map protein fitness landscapes. However, because their training data relies on natural amino acid sequences, these models conflate protein structural constraints with nucleotide mutation biases and codon accessibility. Using the rapid emergence of the divergent influenza A H3N2 K lineage as a stress test, we investigate how base PLMs (ESM-2 and ESM-C) versus fine-tuned versions of these models capture mutational processes. We systematically implement a parameter sweep to explicitly couple (or decouple) empirical nucleotide mutational supply from PLM-assessed amino acid substitution pseudo-probabilities across evolutionary forecasting tasks. We find that base PLMs implicitly learn generic nucleotide-level mutational constraints, an effect strongly amplified by virus-specific fine-tuning. Incorporating explicit mutational accessibility significantly improves the binary prediction of observed amino acid changes. Conversely, when predicting the final circulating frequency of variants that have already emerged, adding mutational supply degrades performance, confirming that selection dominates post-emergence dynamics. Additionally, we perform amino-acid-level epistatic scanning to investigate protein structural constraints in the context of genetic background. This indicates the improbable antigenic substitution I160K is dependent on co-occurring S144N and N158D mutations in the H3N2 K lineage. Ultimately, current PLM pseudo-probabilities are a composite metric that conflates protein structural fitness with historical biases in mutational supply. Explicitly decoupling these independent evolutionary processes optimises predictive accuracy for real-world pathogen forecasting and isolates pure protein fitness for synthetic design pipelines.

evolutionary biology↗

bioRxiv2026

Physics-Driven Zero-Shot Reconstruction of Isotropic 3D Fluorescence Microscopy under Undersampled Acquisition

Three-dimensional (3D) imaging represents the development of next generation of fluorescence microscopy. However, routine axial down-sampling makes isotropic resolution unrealistic. Here, we propose DeepUI, a physical zero-shot framework designed to achieve isotropic 3D fluorescence images from a low axial sampling rate. DeepUI fully leverages the intrinsic characteristics of 3D images through physics-guided degradation, which incorporates spatial-frequency joint learning to generate a scaled optical transfer function, combined with noise degradation and an up-sampling branch. Typically requiring just 5 minutes for training and 0.5 minutes for high-throughput and fast prediction, we demonstrate the superior performance of DeepUI to get isotropic results, and the exclusivity to axial down-sampling conditions, even in more challenging conditions, including defocused background, noise, and resolution blur.

bioinformatics↗

bioRxiv2026

Piezo1 transduces mechanical signals to inhibit osteoclast fusion and coordinate bone homeostasis

Osteoclasts originate from bone marrow-derived macrophages, and their maturation relies on fusion of pre-osteoclasts into mature osteoclasts that maintain bone homeostasis. Previous studies established that mechanical signals regulate osteoblast activity in bone homeostasis, although whether such signals regulate bone homeostasis by acting on osteoclasts remains unclear. Herein, membrane tension decreased during pre-osteoclast fusion, accompanied by reduced Piezo1. This suggests that mechanical stimulation inhibits fusion of monocytes into osteoclasts through Piezo1. Piezo1-knockout monocytes (CTSKCre; Piezo1fl/fl) attenuated the anabolic effect of exercise on bone mass in vivo, whereas shear stress inhibited osteoclast fusion in vitro. Piezo1 activation elevated E-cadherin, which anchors Merlin. This led to hyperactivation of the Yes-associated protein (YAP) signaling pathway, subsequently activating TAX1BP1 and suppressing NF-{kappa}B signaling as well as osteoclast differentiation and maturation. Thus, physical exercise activates a Piezo1-E-cadherin-Merlin-YAP axis that prevents osteoclast fusion, presenting a druggable mechanotransduction pathway for osteoporosis.

cell biology↗

bioRxiv2026

ArchaicSeeker 3.0: A deep-learning framework for scalable, haplotype-resolved inference of archaic introgression

Archaic introgression has left a significant mark on human genetic diversity, but reliably identifying introgressed segments remains a major challenge, especially with complex demographic histories and limited sample sizes. Existing methods often rely on demographic assumptions or cohort-specific parameter fitting, which compromises robustness and scalability. We introduce ArchaicSeeker 3.0 (AS3), a deep-learning framework designed for haplotype-resolved detection of archaic introgression. AS3 integrates a tract-scale sequence model with an overlap-aware reassembly approach and boundary refinement, enabling accurate, boundary-coherent reconstruction of introgressed segments across diverse genomic contexts. By leveraging a simulation-trained model, AS3 avoids inference-time recalibration, offering stable performance across unrepresented demographic scenarios and small cohorts. In extensive simulations, AS3 outperforms existing methods in precision, recall, and F1 score, while providing more continuous segments with accurate boundary localization. It demonstrates robustness in small-target regimes and varying marker densities. Applied to 3,453 genomes from 209 populations, AS3 shows strong concordance with existing introgression callers and identifies additional introgressed regions, including high-frequency AS3-specific introgressed segments supported by locus-level haplotype and phylogenetic analyses. AS3 provides a scalable, robust solution for detecting archaic introgression from single individuals to large biobank datasets, marking a significant advancement in the field of local ancestry inference and opening new possibilities for the study of human evolutionary genetics. ArchaicSeeker 3.0 is available at https://github.com/Shuhua-Group/ArchaicSeeker3.0.

bioinformatics↗

bioRxiv2026

Transcriptomic Profiling of Thyroid Eye Disease Orbital Fibroblasts Identifies Sorafenib as a Novel Therapeutic

Thyroid eye disease (TED) is a debilitating condition characterized by orbital fibroblast (OF) activation and excessive hyaluronic acid (HA) accumulation within the retro-ocular space. While IGF-1R blockade with teprotumumab has significantly advanced TED management, incomplete clinical responses and disease relapse underscore the need to identify alternative targets. In this study, we used high-throughput RNA sequencing to map the transcriptomic landscape in TED OFs compared with non-TED OF controls. Our analysis identified robust enrichment of pathways critical to the TED phenotype, including PI3K-AKT signaling, the platelet-derived growth factor (PDGF) pathway, and extracellular matrix remodeling. We validated several key upregulated mediators that may contribute to orbital remodeling, including FOXC2, HGF, MET, and HMGA2, alongside the downregulation of the Wnt antagonist SFRP2. By employing a computational drug-repositioning approach, we identified the multi-kinase inhibitor sorafenib, which targets VEGFR, PDGFR, and RAF, as a potent candidate to neutralize the TED-specific gene signature. Functional assays demonstrated that sorafenib dose-dependently inhibited PDGF-induced AKT phosphorylation and significantly attenuated HA synthesis in primary TED OFs. These results define a persistent, receptor tyrosine kinase-driven program in the TED orbit and suggest that multi-kinase inhibition represents a viable therapeutic strategy for refractory TED. HighlightsO_LIThyroid eye disease (TED) orbital fibroblasts exhibit a transcriptomic signature characterized by elevated PI3K/AKT, angiogenic, and growth factor signaling. C_LIO_LIComputational drug prediction identifies sorafenib as a candidate to reverse the TED gene signature. C_LIO_LISorafenib dose-dependently inhibits AKT activation and hyaluronic acid production in TED orbital fibroblasts. C_LI

molecular biology↗

bioRxiv2026

Explainable Physicochemical Determinants of Protein Ligand Binding via Non-Covalent Interactions

Protein-ligand binding governs enzymatic catalysis, metabolic homeostasis, and therapeutic modulation. Thus, the accurate prediction of these interactions underpins modern rational drug discovery. However, existing deep-learning frameworks largely operate as black-box predictors that fail to resolve the individual residues mediating binding or decode the fundamental non-covalent forces that drive molecular recognition. To address these limitations, we present ExplainBind, an interaction-aware framework that predicts binding likelihood, localizes specific binding residues at single-amino-acid resolution rather than coarse pocket-level regions, and decodes the underlying non-covalent interaction patterns, all zero-shot, without requiring prior three-dimensional structural inputs. To support residue- and interaction-level training and evaluation, we construct InteractBind, a protein-ligand benchmark with residue-atom interaction maps. Simultaneously, benchmarking experiments demonstrate that ExplainBind consistently outperforms state-of-the-art baselines across diverse protein and ligand spaces, maintaining high precision when generalized to entirely novel sequences and chemical scaffolds. When applied to two unseen therapeutic targets, ExplainBind successfully ranks potent angiotensin-converting enzyme (ACE) inhibitors and clarifies differences in their potency via affinity-stratified interaction landscapes. Furthermore, we demonstrate the prospective utility of ExplainBind by discovering novel inhibitors and activators of L-2-hydroxyglutarate dehydrogenase (L2HGDH) through wet-lab validation, with mechanistically distinct interaction profiles providing a clear molecular rationale for their divergent functional outcomes. Collectively, these results establish ExplainBind as a powerful, generalizable tool for mechanistically informed, interpretable drug discovery.

molecular biology↗

bioRxiv2026

Morphospatial profiling of cancer-associated fibroblasts reveals architectural subtypes of pancreatic ductal adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) is a lethal malignancy with an urgent need for biomarkers to predict prognosis and guide treatment. Understanding the complex spatial biology of pancreatic cancer-associated fibroblasts (CAFs) and the broader architecture of the PDAC tumour microenvironment is central to this challenge. Using a multi-omics approach across multiple spatial resolutions in a large human PDAC cohort, we integrate geometry and shape to define discrete morphological CAF subtypes, expanding CAF phenotyping beyond conventional proteomics. We then reveal an architectural and molecular axis of PDAC at tissue level, suggestive of epithelial-stromal co-evolution, with translational implications and prioritisation of stromal targets. Finally, we recapitulate this axis by introducing four unique, internally validated architectural subtypes of PDAC, each characterised by a common microenvironment and CAF enrichment profile. These archetypes outperform conventional pathology in prognostication, and predict response to adjuvant chemotherapy. Collectively, this study establishes a novel morphological paradigm for spatial biology, illuminates the architectural landscape of PDAC, and provides a framework for spatial biomarker discovery to close the translational gap in this devastating disease.

cancer biology↗

bioRxiv2026

Asymmetric partition of the O-GlcNAcome in mitosis ensures binary cell fate decision

Many cellular components undergo biased segregation during stem cell division, but whether such order extends to the proteome has remained unknown. Using our real-time O-GlycoTracer, we find that O-GlcNAcylated proteins (the O-GlcNAcome) segregate asymmetrically during Drosophila neuroblast mitosis, predisposing the two daughters to distinct fates. The daughter that preserves stem cell identity inherits most of the O-GlcNAcome. This asymmetric partition requires putative O-GlcNAc readers, notably 14-3-3 proteins. We further identify the nuclear pore complex (NPC) as a major O-GlcNAc substrate in neuroblasts; Nup153, located in the nuclear basket, shows O-GlcNAc-dependent biased inheritance. Perturbing multiple steps in this segregation pathway disrupts neuroblast differentiation, reduces brain size, and causes adult learning deficits. These findings reveal a coordinated O-GlcNAc-driven mechanism for proteome-level asymmetric inheritance during neural stem cell division, with implications for neurodevelopmental disorders linked to OGT or 14-3-3 mutations.

cell biology↗

bioRxiv2025

The Overlap Area as a Novel Measure of Effect Size in Neuroscience Research

In experimental biomedical research, a common concern is whether a manipulation produces a biologically meaningful effect. Another concern is whether effect sizes are reliable, statistically significant, and generalizable. Traditional effect size measures, such as Cohens d, quantify mean differences but ignore variance heterogeneity between groups. This can result in biased effect size estimates and a lack of thresholds for statistical significance testing. Motivated by this, we introduce a novel effect size measure, termed overlap area (OA), which quantifies the difference between the population distributions of the experimental and control groups. A robust Bayesian method estimated OA, and random resampling determined OA thresholds for statistical significance in single and replicated experiments. Simulations confirmed the approachs sensitivity and robustness. Applied to a real-world dataset, OA revealed that environmental enrichment affects the mouse brain synaptome. Moreover, we developed an open-source toolbox supporting OA as a powerful new measurement for conducting robust, reliable, and reproducible analyses of manipulation effects in neuroscience and related fields.

neuroscience↗

bioRxiv2025

Experiences are encoded by brainwide reprogramming of synaptome architecture

Synaptome architecture describes the spatiotemporal distribution of highly diverse excitatory synapses throughout the brain. Whether and how this architecture is impacted by experience is key to understanding its role in learning and memory. We found that environmental enrichment and monocular visual deprivation drive large-scale, type-and subtype-specific reorganisation of excitatory synapses in more than one hundred brain regions. Each experience modifies distinct subsets of synapses, with patterns aligned with protein turnover rates and connectome architecture. These reorganisations occur during development and adulthood, revealing a conserved mechanism of synaptome plasticity across the lifespan. Our findings support a population-selection model in which experience drives adaptation by selectively modifying synapse varieties, generating a distributed trace of past experiences. Our results also point to synaptome architecture as a shared framework integrating experience, lifespan changes, sleep, genetic variation and disease.

neuroscience↗

bioRxiv2025

Self-supervised AI reveals a lethal discohesive phenotype in lung adenocarcinoma

Applications of artificial intelligence (AI) to histopathology are now common, but most require supervision which inherently limits their scope. By using self-supervised learning (SSL), we discover and quantify the full range of histopathological appearances in a disease, and associate them with clinicopathological ground truths such as prognosis. We used this approach to discover under-appreciated morphologies of lung adenocarcinoma (LUAD), using a highly characterised resected tumour cohort of over 4000 slides from over 1000 patients. By constructing an authoritative lexicon of recurrent LUAD appearances, we ab initio discovered several stromal morphologies strongly predictive of outcome. With multimodal data integration and external dataset validation, we propose that epithelial discohesion is lethal, but only in the context of immunologically cold stroma. Both these morphological features are independent of current prognostic schema. Crucially, we describe these features in the context of real-world diagnostic histopathology, giving them immediate clinical translatability. Statement of SignificanceHistopathology relies heavily on epithelial morphology, often neglecting the stroma. We use self-supervised AI to identify under-appreciated morphologies in LUAD linked to poor outcome and validate these observations in an external cohort, demonstrating the utility of self-supervised AI as a powerful biological discovery tool.

cancer biology↗

bioRxiv2025

Self-Supervised AI Reveals a Hidden Landscape of Prognostic Spatial Patterns in Multiplex Immunofluorescence Images

Modern spatial proteomic methods, such as multiplex immunofluorescence (mIF) imaging, offer a data-rich view of spatial biology in intact tissues. However, interpreting its complexity is a major bottleneck, limiting its potential for biological discovery and clinical translation. Current computational methods often rely on segmentation-based approaches that discard crucial morphological information and are limited to testing pre-defined hypotheses. Here, we introduce a self-supervised learning (SSL) framework that enables hypothesis-agnostic, context-aware discovery of biomarkers directly from mIF images. Our approach extracts rich feature representations that capture holistic architectural patterns, which integrate cellular morphology, marker interactions, and microenvironmental context without human supervision. Applying this framework to over 7,000 mIF tissue images from over 1,800 patients in two distinct cancer types, we demonstrate superior prognostic performance over conventional segmentation analyses. The method autonomously identified previously unknown and potentially clinically actionable biological patterns. In lung adenocarcinoma, these include a Ki67-mediated immune evasion phenotype, a sub-cellular pattern of GLB1 expression which aligns with low-grade EGFR-driven tumours, and distinct modes of tumour-immune interaction in PD-L1+ patients. We also find a regulatory T-cell-mediated immunosupressive environment promoting tumour budding in colorectal carcinoma. Our work establishes SSL as a powerful, scalable, and unbiased platform to decode tissue ecosystems while being fully explainable without pre-defined hypotheses. This paradigm shift transforms high-plex imaging from a hypothesistesting tool into a hypothesis-generating engine that can accelerate the discovery of next-generation spatial biomarkers.

cancer biology↗

bioRxiv2025

PRMix: Primary Region Mix Augmentation and Benchmark Dataset for Precise Whole Mouse Brain Anatomical Delineation

The architecture of the mouse brain shares remarkable similarities with the human brain, making it an essential model for studying brain pathologies, synaptic diversity, and regional specialization. A key step in such studies involves registering molecular images to reference brain atlases, a process hindered by the difficulty of accurately delineating brain regions. Toward this, we have curated a collection of high-resolution, dual-fluorescence microscopy images, termed as dual-fluorescence mouse brain microscopy (DMBM) dataset, complemented by expert annotations of 118 subregions. This dataset provides unprecedented insights into the molecular and structural complexity of the mouse brain. However, its full potential for detailed whole-brain analysis is compromised by challenges such as boundary ambiguity and sample scarcity in existing automated segmentation methods, prompting the development of the primary region mix (PRMix) augmentation method. PRMix is specifically designed to expand these datasets, enhance the realism of synthetic data and minimize overlap between adjacent regions. Our approach, together with the curated dataset, achieves superior segmentation performance across the mouse brain compared with existing methods, setting a new benchmark in brain imaging research. Code and data are available at https://git-pages.ecdf.ed.ac.uk/dmbm-datasets-5c13cd/.

neuroscience↗

bioRxiv2024

A histomorphological atlas of resected mesothelioma from 3446 whole-slide images discovered by self-supervised learning

1Mesothelioma is a highly lethal and poorly biologically understood disease which presents diagnostic challenges due to its morphological complexity. This study uses self-supervised AI (Artificial Intelligence) to map the histomorphological landscape of the disease. The resulting atlas consists of recurrent patterns identified from 3446 Hematoxylin and Eosin (H&E) stained images scanned from resected tumour slides. These patterns generate highly interpretable predictions, achieving state-of-the-art performance with 0.65 concordance index (c-index) for outcomes and 88% AUC in subtyping. Their clinical relevance is endorsed by comprehensive human pathological assessment. Furthermore, we characterise the molecular underpinnings of these diverse, meaningful, predictive patterns. Our approach both improves diagnosis and deepens our understanding of mesothelioma biology, highlighting the power of this self-learning method in clinical applications and scientific discovery.

pathology↗