Biology subjects

Nguyen, T. H.

Publications and source records attributed to Nguyen, T. H..

At least 19 recordsLinked to original sources

bioRxiv2026

A broadly active flavivirus NS4B targeting compound that blocks NS4B-NS3 binding and replication organelle formation

Mosquito-borne orthoflaviviruses, including West Nile virus (WNV), dengue virus (DENV), Zika virus (ZIKV), and Japanese encephalitis virus (JEV), cause substantial global morbidity and mortality, yet no broadly effective direct-acting antivirals are available. The viral nonstructural protein NS4B has emerged as a promising antiviral target, although the compounds described to date are largely DENV-specific. Here we characterize SRI-38841, a xanthine analog derived from hit-to-lead optimization of a previously described WNV inhibitor, which inhibits WNV replication with an IC90 of 0.87 {micro}M and low cytotoxicity, and retains activity against DENV, ZIKV, and JEV. Serial passage of WNV under compound selection yielded resistant viruses bearing mutations in NS4B, and reverse genetics was used to confirm the role of individual amino acid substitutions. Several of these residues map to regions analogous to those implicated in resistance to the DENV NS4B inhibitors JNJ-A07 and NITD-688. Docking of SRI-38841 to an AlphaFold2 model of WNV NS4B predicted a binding pocket within a cytosolic-facing funnel, and mutation of the pocket residues A45 and K56 likewise conferred resistance, while changes at Q170 and T220 did not. Together, these data favor a topology model in which the compound engages a cytosolic-facing pocket over one placing these residues in the ER lumen. Mechanistically, SRI-38841 disrupts the NS4B-NS3 interaction and abolishes replication organelle formation in infected cells. Unlike JNJ-A07 and NITD-688, which were inactive against WNV and ZIKV, SRI-38841 inhibits all four viruses tested, identifying this xanthine class as a starting point for broadly active orthoflavivirus antivirals.

microbiology↗

bioRxiv2026

Caenorhabditis becei recombinant inbred lines (beRILs) reveal the scope of heritable variation within a gonochoristic nematode population.

Caenorhabditis nematodes are a powerful model clade for evolutionary genetics. Isogenic lines and panels of recombinant inbred lines (RILs) are among the most essential tools for genetic studies in these species. While most Caenorhabditis species are gonochoristic, large RIL panels have only been developed for self-fertilizing species. This gap biases our understanding and limits our ability to address questions related to the genetic architecture of traits in outbred populations, which have radically higher genetic diversity, heterozygosity, and effective recombination than selfers. Having previously identified Caenorhabditis becei as a tractable gonochoristic species due to its moderate inbreeding depression, we generated two panels of advanced-intercross RILs derived from three individual outbred C. becei worms collected from a single locality on Barro Colorado Island, Panama. One panel derives from a pair of worms sampled from a single rotting fig; the other derives from a cross between worms from two different figs. The panels share one founder in common, yielding two half-sib RIL panels. We sequenced and haplotyped the lines, identifying millions of variants and thousands of recombination breakpoints. Using simulations, we demonstrate the suitability of these lines for quantitative genetics studies and QTL mapping. In our single-fig panel, we observe abundant heritable variation in population growth rate, individual body size, and sexual dimorphism for body size. We detected four QTLs associated with population growth rate and show that estimated allelic effects are good predictors of selection that occurred during panel derivation.

genetics↗

bioRxiv2026

Phage-based microbiome manipulation reveals ecological interactions within gut communities

Mechanistic understanding of gut ecology is limited by the availability of tools for precise manipulation of microbiome composition. Here, we isolate lytic phages to enable targeted removal of gut commensal Escherichia fergusonii (Ef) from complex, undefined stool-derived in vitro communities. A single phage drove resistance without fitness cost in monoculture, but resistant Ef exhibited reduced fitness in communities, enabling expansion of closely related Proteobacteria. Resistance arose via reversible promoter inversion linked to outer-membrane function. A phage cocktail overcame resistance to achieve Ef knockout across communities with minimal collateral effects. Using knockout communities, we show that Ef is necessary and sufficient for preventing Salmonella invasion. Replacement with an Ef transposon-mutant library revealed that community-specific fitness defects are enriched in genes involved in outer-membrane assembly. Disruption of these genes sensitized Ef to antagonistic community members, highlighting interspecies warfare as a key driver of microbiome ecology. These results establish phage-mediated perturbation as a framework for linking species to community-level function and for enabling precision microbiome engineering.

microbiology↗

bioRxiv2026

STRADA Deficiency Impairs Cortical Interneuron Development in Humans and Mice

Polyhydramnios, Megalencephaly, and Symptomatic Epilepsy syndrome (PMSE/STRADA-related disorder) is a rare neurodevelopmental disorder characterized by megalencephaly (ME), early-onset drug-resistant epilepsy, neurocognitive impairment, and high early mortality, often due to status epilepticus. PMSE is caused by a multi-exon deletion in STRADA, encoding STRADA, which regulates the mechanistic target of rapamycin (mTOR) pathway. GABAergic inhibitory interneurons (INs) critically modulate the excitatory:inhibitory balance in cortical and hippocampal networks, and IN deficits contribute to epileptogenesis in several epileptic encephalopathies. However, no studies have investigated INs in PMSE. We used a multimodal approach to study INs in a Strada-/- mouse model engineered with the same causative 5-exon deletion identified in human PMSE. We demonstrate that Strada/STRADA loss causes a reduction of INs in the somatosensory cortex and a corresponding increase in the striatum, representative of remnant ganglionic eminence progenitor origin, in Strada-/- mice and a single PMSE brain tissue specimen. RNA sequencing comparing wildtype to Strada-/- cortex and striatum corroborated these findings, revealing increased IN-related gene expression (e.g., Dlx2) in the striatum and decreased IN-related gene expression (e.g., Pvalb) in the developing cortex. Cytoskeletal (e.g., Tpp3, Kank4, Map1a) and mTOR-associated genes (e.g., Rictor, Cryab) are differentially expressed in the developing cortex, mature striatum, and mature cortex of Strada-/- mice. Functional validation confirmed enlarged INs in mouse and human Strada/STRADA-deficient brain and enhanced S6 phosphorylation in Strada-/- striatum. Together, these findings suggest STRADA/Strada loss contributes to failed IN migration -- the first such report in a developmental, mTOR-associated megalencephaly syndrome -- highlighting INs as a therapeutic target for seizure prevention in PMSE. Key PointsO_LI- Reduced numbers of cortical inhibitory interneurons were observed in the cerebral cortex of Strada-/- mice, with striatal interneuron aggregation C_LIO_LI- Reduced numbers of cortical inhibitory interneurons, with an aggregation in striatum, were observed in human PMSE brain, supporting the observations in Strada-/- mouse C_LIO_LI- Transcriptomic analysis in Strada-/- mice reveals evidence of early developmental interneuron and cytoskeletal dysfunction C_LIO_LI- We introduce a loss of cortical interneurons as a salient feature of PMSE developmental pathogenesis, potentially contributing to a loss of inhibitory modulation C_LIO_LI- This is the first study proposing interneuron migration impairment in the developmental pathogenesis of an mTOR-associated megalencephaly syndrome C_LI

neuroscience↗

bioRxiv2026

Disulphide and sequence-encoded conformational priors guide nanobody structure prediction

Nanobody binding is largely governed by the HCDR3 loop, which adopts distinct placement regimes relative to the framework: compact, framework-contacting (kinked blueprint) and solvent-exposed (extended blueprint). Many nanobodies also contain additional cysteines that form non-canonical disulphide bonds, imposing covalent constraints on binding-loop conformations. Current structure predictors are typically trained and benchmarked with smooth coordinate-based objectives, so models may appear reasonable under root-mean-square deviation (RMSD), while adopting an incorrect HCDR3 blueprint or failing to recover the native disulphide connectivity, impacting paratope geometry and functional interpretation. Here, we show that the HCDR3 blueprint is predictable from sequence alone, allowing for explicit constraints during modelling. We implement these principles into NbForge, a lightweight nanobody folding model that incorporates blueprint- and disulphide-aware inductive biases and is trained with filtered self-distillation. NbForge improves recovery of HCDR3 blueprint and non-canonical disulphide formation over previous lightweight models and achieves coordinate accuracy at par to state-of-the-art, large, resource-intensive predictors, while running at sub-second inference speed. We show that using NbForge monomer models as templates further improves the success rate of predicting nanobody-antigen complexes. Together, these results motivate blueprint- and disulphide-aware benchmarks for nanobody modelling beyond RMSD, and show that appropriate inductive biases can close the performance gap to heavyweight predictors. We make the sequence classifier (NbFrame) and NbForge available for download and via a user-friendly web server.

bioinformatics↗

bioRxiv2026

Structural probing of RNA hairpins quantifies protein occupancy on RNA and links it to function in human cells

RNA structures that form inside living cells influence processes ranging from translation to RNA decay, many of which are controlled by RNA-binding proteins (RBPs). Because RBP specificity depends on both local RNA structure and sequence motifs, traditional pulldown-based methods often obscure the structural context of bound RNAs. Here, a quantitative framework based on dimethyl sulfate mutational profiling and sequencing (DMS-MaPseq) is introduced that jointly measures RNA structure and protein binding at single-nucleotide resolution in human cells, enabling estimation of effective RNA-protein affinities and fractional occupancy directly in cells. Application of this approach to a 1,600-member library of MS2 hairpin mutants reveals that RNA folding is the strongest determinant of MS2 coat protein (MCP) recognition, and that stable MS2 structures must form for MCP to bind its target sequence. MCP also shows strong preference for its consensus loop sequence while displaying minimal dependence on stem length beyond ten base pairs or on stem GC content. Incorporation of an inducible degron fused to MCP allows precise tuning of intracellular protein concentrations analogous to those of many endogenous RBPs and show that DMS reactivity changes can be used to infer binding specificities across a subsaturating regime. A quantitative occupancy framework further shows that inferred fraction-bound values accurately predict how efficiently MCP fused to a downregulatory-domain drives RNA degradation. Together, these results establish a generalizable approach for measuring RBP-RNA affinities with structural resolution in living cells, dissecting how sequence and structure contribute to RBP recognition, and quantitatively linking occupancy to functional output.

biochemistry↗

bioRxiv2026

HvbZIP33 and HvbZIP76 have overlapping roles in foliar transpiration in drought-stressed barley

Basic leucine zippers (bZIP) constitute one of the biggest protein families and evolutionarily conserved transcription factors (TFs) in plants. We obtained mutant lines for two bZIP TFs, HvbZIP33 and HvbZIP76, in the genetic background of the barley cultivar Golden Promise (GP) via targeted gene-specific mutagenesis using an RNA-guided Cas9 endonuclease. A comprehensive morphological, physiological and transcriptomic analysis was performed in wild-type GP compared with hvbzip33 and hvbzip76 mutants under drought stress. The morphological and physiological changes were similar in both mutants and in the wild-type GP. Most strikingly, the mutants exhibited accelerated wilting and increased water loss. This effect was primarily caused by higher stomatal conductance (gs) and transpiration rate (E) in mutants compared to wild-type GP under both control and drought conditions, which in turn had a detrimental effect on the mutants intrinsic water use efficiency (iWUE). Likewise, the transcriptome profiles of hvbzip33 and hvbzip76 were more similar to each other than those of wild-type GP. We found that the number of differentially regulated genes under control versus drought-stress conditions was higher in the mutants than in wild-type GP, suggesting that the mutants try to compensate for accelerated foliar water loss. The study highlights the essential roles of HvbZIP33 and HvbZIP76 in balancing water loss in barley. These findings provide a foundation for engineering enhanced drought tolerance in barley through targeted manipulation of these genes to optimize transpiration rates.

genetics↗

bioRxiv2025

SurfDiff: protein surface profiling for selective or broadly reactive epitope prioritisation in binder and immunogen design

Rational selection of protein-surface patches - epitopes or ligandable sites - is essential for developing targeted antibodies, binding proteins, peptides, or small-molecule ligands, and for informing immunogen design in vaccines. We present SurfDiff, a structure- and sequence-informed framework for protein surface comparison. SurfDiff supports one-to-one and one-to-many comparisons by combining local structural alignments with physicochemically and spatially aware neighbourhood analysis. It assigns residue-level uniqueness and similarity scores, which can be aggregated into surface-level similarity, selectivity, and a discriminability score that captures conservation across desired targets while penalising similarity to undesired off-targets. These scores, calculated solely from the targets without any knowledge of the binders, reliably predict known experimental selectivity for diverse binders including small molecules, peptides, antibodies and antibody mimetics across viral antigens, cytokine isoforms, GPCR subtypes and serum albumin. They also correlate strongly with binding and neutralisation data across pathogen variants, and substantially outperform commonly used bioinformatic metrics such as sequence substitution matrices and structural similarity measures. SurfDiff provides a generalisable and interpretable approach to protein surface profiling, enabling selective or cross-reactive binder design, cross-species prioritisation, and facilitating immunogen selection. We make SurfDiff available as downloadable open-source software: (https://gitlab.developers.cam.ac.uk/ch/sormanni/surfdiff) and as a webserver: www-cohsoftware.ch.cam.ac.uk/index.php/surfdiff.

bioinformatics↗

bioRxiv2025

Cereblon Promotes Influenza Virus Replication through AMPK Ubiquitination

Influenza A and B viruses (IAV and IBV) continually threaten global health, with IAV posing a risk of emerging pandemics. Rapid viral evolution makes current treatments less effective, highlighting the urgent need for broad-spectrum antivirals. Targeting host factors essential for viral replication may offer a highly promising broad-spectrum antiviral strategy. In this context, cereblon (CRBN), a substrate adaptor of the CRL4 E3 ubiquitin ligase complex, is found to promote both IAV and IBV replication as a key pro-viral host factor. Mechanistically, CRBN targets and degrades AMP-activated protein kinase (AMPK) via the proteasome. This CRBN-driven degradation shifts the metabolism of the infected cells toward anabolism, promoting lipid droplet (LD) formation and creating a microenvironment favorable for viral replication. Genetic depletion or inhibition of CRBN stabilizes AMPK, significantly reduces LD formation, and effectively suppresses the replication of various IAV and IBV strains in vivo, demonstrating its broad-spectrum potential. Notably, Crbn knockout mice show marked resistance to lethal IAV infection. CRBN inhibition with immunomodulatory imide drugs--known CRBN inhibitors--significantly decreases IAV replication in vivo. This research underscores CRBN as a crucial regulator of host metabolism during viral infection, revealing the CRBN-AMPK axis as a promising target for host-directed pan-influenza antiviral development. Author SummaryInfluenza A and B viruses are response for recurring seasonal epidemics and occasional pandemics that pose serious global health challenges. Current antiviral drugs often lose effectiveness as influenza viruses rapidly develop resistance through genetic mutations. To overcome this limitation, our study focused on a host factor targeted by influenza viruses-the ubiquitin E3 ligase substrate adaptor cereblon (CRBN). We discovered that influenza viruses exploit CRBN to enhance their replication by altering host cell metabolism. Specifically, CRBN binds to a key energy sensor - AMPK and induces the ubiquitination and degradation of AMPK{gamma}. Consequently, the AMPK suppression shifts the cellular environment from catabolism to anabolism, activating lipogenic enzymes and promoting lipid droplets formation which provide a favorable environment for viral growth. Our study highlight that blocking CRBN either genetically or chemical inhibition (immunomodulatory imide drugs) of CRBN reduces LD formation and strongly suppresses diverse influenza strains replication in vitro and in vivo. These findings reveal not only how viruses hijack host metabolism via the CRBN-AMPK pathway but also present a potential therapeutic for novel broad-spectrum anti-influenza.

microbiology↗

bioRxiv2025

Integron-associated structures underlie the accumulation of antibiotic resistance genes on plasmids beyond cassette arrays

Rising multidrug-resistant (MDR) bacteria threatens global public health. While antibiotic selection preserves MDR plasmids, how mobile genetic elements (MGEs) drive the antibiotic resistance gene (ARG) accumulation to form complex plasmid architectures remain elusive. Here, we demonstrate that MDR plasmid evolution follows a hierarchical scheme, where progressive combinations of integrons, IS26, and ISCR elements drive stepwise increases in ARG load. The combination of all these three MGEs were significantly linked to the highest ARG load on plasmids. In the presence of integrons or ISCR elements, ARG numbers in plasmids are strongly and positively correlated with IS26 copy numbers. Such correlations were further validated by Escherichia coli plasmids collected from human and pig wastewater. We propose an integron-IS26-ISCR (3I) framework with four MDR plasmid evolutionary stages of increasing ARG accumulation potential. The "3I" framework refines predictions of MDR plasmid emergence and spread, providing universal genetic markers for cost-effective surveillance.

microbiology↗

bioRxiv2025

Environmental and vector detection of African swine fever virus DNA in Vietnam: Evidence for potential transmission through wastewater and Amblyomma javanense ticks

African swine fever virus (ASFV) remains a major global threat to pig production, yet its environmental and vector reservoirs in Southeast Asia are poorly understood. This study provides the first evidence in Southeast Asia indicating that ASFV DNA can be detected in pig farm wastewater and that Amblyomma javanense ticks may act as potential mechanical carriers of the virus in natural settings. To elucidate part of the environmental dynamics and possible transmission routes of ASFV, we developed and applied a highly sensitive polyethylene glycol (PEG) precipitation-based real-time PCR method specifically for wastewater surveillance, as the viral load in aquatic environments is presumed to be substantially lower than that in arthropod vectors. A total of 40 tick samples and 93 wastewater samples were analyzed. ASFV DNA was identified in 3 of 40 ticks (7.5%) and 7 of 93 wastewater samples (7.5%). The developed method showed a 100-fold improvement in analytical sensitivity, lowering Ct values by 3-7 cycles and increasing the detection rate from 3.2% to 7.5% compared with the reference method. Diagnostic sensitivity and specificity for wastewater testing were 100.0% and 95.6%, respectively. Although infectious virus was not recovered from ASFV-positive wastewater samples, control experiments confirmed that PEG treatment did not interfere with viral infectivity. These findings highlight the potential role of environmental and tick-mediated transmission in ASF epidemiology and provide a practical tool for early warning and post-outbreak monitoring of ASFV in endemic regions.

microbiology↗

bioRxiv2025

Target-enriched metagenomics-informed qPCR detects rare, potentially dangerous β-lactamase genes in wastewater

The rapid emergence of novel antibiotic resistance genes (ARGs) decreases the effectiveness of empirical antibiotic treatment for pathogen infections. Environmental ARG surveillance is an early-warning approach that can better inform antibiotic usage. Quantitative polymerase chain reaction (qPCR) is widely used for ARG surveillance because qPCR is easy, fast, and highly sensitive. However, it can only identify DNA targeted by pre-selected primers. In contrast, metagenomic sequencing can identify ARGs agnostically. However, sequencing is more expensive, less sensitive, and requires lengthy analysis of complex data sets. Target-enrichment metagenomic sequencing (TEMS), a method we developed previously, can mitigate the disadvantage of low sensitivity of traditional metagenomic sequencing. In this study, we propose a hybrid ARG surveillance pipeline for wastewater that capitalizes on both TEMS and qPCR. It uses a large-scale target determination of ARGs by using TEMS, followed by fine-scale routine surveillance of ARGs using qPCR. To connect the two steps, we developed a primer design tool, MSEDAP, which automatically analyzes metagenomic sequencing data and designs qPCR primers for ARG surveillance. This new workflow was ground-truthed using wastewater samples. TEMS identified seventeen {beta}-lactamase gene targets of potential clinical importance. qPCR validated their presence and abundance using primers generated by MSEDAP. SynopsisA metagenomics-qPCR hybrid scheme can be used for environmental surveillance of antibiotic resistance genes (ARGs) to discover emerging ARGs originating from communities. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=110 SRC="FIGDIR/small/665820v1_ufig1.gif" ALT="Figure 1"> View larger version (29K): org.highwire.dtl.DTLVardef@625e06org.highwire.dtl.DTLVardef@ae34aorg.highwire.dtl.DTLVardef@d03e7borg.highwire.dtl.DTLVardef@1b625a9_HPS_FORMAT_FIGEXP M_FIG C_FIG

microbiology↗

bioRxiv2025

Maternal diet exerts sex-specific effects on offspring' personalities in predatory mites

Animal personality is characterized by consistent behavior within individuals linked to consistently variable behavior among individuals in a population across time and contexts. Genetic determination, transgenerational effects, and personal experience are major pathways shaping animal personalities. Among these pathways, little attention has been paid to environmental factors in the parental generation affecting offspring personality. Here we tested the effects of the maternal diet on offspring personality in the plant-inhabiting predatory mite Amblyseius swirskii. Mated females and males, whose mothers were fed during egg production on either cattail pollen, two-spotted spider mites, or thrips, were subjected to a battery of three to five tests each for exploration, activity, and boldness. Movement activity was assessed in the mites familiar environment. Exploration was quantified by the latency to leave and reach novel sites or objects. Boldness was evaluated by residence in risky and benign sites. Mean behaviors were analyzed by generalized estimating equations, repeatability was assessed by intraclass correlation coefficients. On average, offspring from spider mite-fed mothers were the most active and those from pollen-fed mothers were the shiest. Offspring from thrips-fed mothers were more repeatable in activity than offspring from pollen-and spider mite-fed mothers. Consistently little and highly active personalities produced more eggs than inconsistent, flexible types. Only offspring from pollen-fed mothers were repeatable in boldness. Maternal diet did not influence offspring personality in exploration. Taken together, our study suggests that the maternal diet critically influences both mean behavioral trait expression and behavioral repeatability of offspring. The ability of mothers to respond to short-term diet changes during internal egg production allows to adaptively adjust the behavior and personalities of daughters and sons within local and regional groups.

animal behavior and cognition↗

bioRxiv2025

Fusion transcription factor dosage controls cell state in rhabdomyosarcoma

In the fusion-positive subset of rhabdomyosarcoma, the PAX3::FOXO1 oncoprotein is the most common fusion driver. We previously established a human myoblast system for inducible expression of PAX3::FOXO1. In the current study, we modulate PAX3::FOXO1 protein expression to understand the epigenetic and phenotypic functions at different PAX3::FOXO1 levels. Proliferative and oncogenic outcomes depend on PAX3::FOXO1 dosage in this system with transformation dominant at intermediate levels and growth suppression dominant at high levels. After prolonged PAX3::FOXO1 expression, there is dosage-dependent heterogeneity in single cell gene expression profiles. We observe a dosage-specific effect for PAX3::FOXO1 chromatin recognition and identify factors that modulate PAX3::FOXO1 chromatin binding. PAX3::FOXO1 dosage affects expression signatures related to cell cycle, epithelial-mesenchymal transition, and myogenesis. Whereas intermediate PAX3::FOXO1 expression maximizes chromatin binding to modulate gene expression, high PAX3::FOXO1 expression alters S phase progression and increases accessibility behind the replication fork. We conclude that PAX3::FOXO1 exerts dosage-dependent functions to influence epigenetic heterogeneity in fusion-positive rhabdomyosarcoma.

genomics↗

bioRxiv2025

Abundant Recurrent Mitochondrial Mutations and Widespread Mitonuclear Epistasis in Caenorhabditis elegans

Coordinated genetic and physical interactions between mitochondrial and nuclear gene products regulate ATP production in the mitochondria. Linking mitochondrial genotypes and mitonuclear genetic interactions to phenotypes remains a complex challenge. Here, we have developed Caenorhabditis elegans as a model for mitonuclear epistasis studies. In a sample of 540 genetically distinct wild isolates, 10% of sites in the mitochondrial genome vary, with hundreds of missense mutations segregating in the species. Recurrent mutations and triallelic sites are common. Phylogenetic analyses of mitogenome sequences identified eight distinct lineages, each with diagnostic variants. Principal component analysis of the nuclear genomes showed considerable concordance between mitochondrial and nuclear genomes in C. elegans populations, suggesting that disrupting coevolved mitonuclear genetic combinations could reveal substantial epistasis. We used GPR-1 overexpression, which disrupts the first mitotic division, to efficiently exchange nuclear and mitochondrial genomes between all pairs of 18 naturally isolated C. elegans strains, generating the largest-to-date animal mitonuclear exchange panel, with 323 unique viable mitonuclear genotypes. We phenotyped development of a subset of strains, with 30 unique genotypes, under six different environmental conditions, including high temperature and exposure to heavy metals. Mitonuclear epistasis contributed significantly to phenotypic variance across all tested conditions. We also tested for mitonuclear coadaptation by comparing the stress resistance of matched and mismatched cybrids. Interestingly, some mismatched strains exhibited greater resistance, highlighting the complexity and context dependence of mitonuclear interactions.

evolutionary biology↗

bioRxiv2025

Autonomous learning of pathologists' cancer grading rules

Deep learning reveals that tissue morphology contains rich pathophysiological information beyond human understanding. However, approaches to convert these spatially distributed signals into subcellular insights informing disease mechanisms are lacking. We introduce Delta-Marches, an interpretability-first approach that nominates distinguishing morphological features rather than explaining existing models' decisions. Delta-Marches simulates idealized morphological changes between classes by coupling latent-space traversals with generative AI. Comparing each image to its class-shifted counterpart allows feature extractors to infer the most affected aspects, reducing sample-to-sample variability and resolving transformations at subcellular resolution. Prototyped in renal carcinoma grading, Delta-Marches generates realistic grade transitions and pinpoints tumor-cell nuclear phenotypes as key determinants. It also reveals reduced vasculature with increasing grade, a known pattern absent from standard rubrics. Applied to dysplastic progression in colorectal tissue, it recovers glandular remodeling and goblet-cell loss, generalizing across tissue types and scales. These results show Delta-Marches parses complex image phenotypes and catalyzes hypothesis generation.

bioinformatics↗

bioRxiv2025

Quantification of collagen and associated features from H&E-stained whole slide pathology images across cancer types using a physics-based deep learning model

BackgroundCollagen is the major component of the extracellular matrix (ECM). Collagen structural organization undergoes significant transformation during tumorigenesis. The visualization of collagen in histological tissue sections would aid in the study of tumor growth, encapsulation, and invasion. However, such visualization requires the use of special stains such as Picrosirius Red (PSR) or Massons Trichrome (MT), or more recently, second-harmonic generation imaging (SHG) in unstained tissue sections. However, PSR and MT both suffer from significant inter- (and intra-) lab stain variabilities, and SHG, while considered a ground truth by many, suffers from issues of system complexity/reliability, cost, and speed/throughput. These technical hurdles limit more widespread assessment of collagen in tissue samples. MethodsUsing high-contrast, high-throughput polarization imaging on PSR-stained slides to generate ground truth training polarization images, we developed a deep learning model (iQMAI) to infer the presence of collagen directly from hematoxylin and eosin (H&E)-stained whole-slide images (WSIs) with high specificity. After iQMAI inference across WSIs, individual collagen fibers were extracted, and features describing overall collagen intensity and fiber morphology were computed. iQMAI pixel-and feature-wise outputs were compared to ground truth polarization imaging to assess model performance. The trained iQMAI model was deployed on H&E-stained WSI from the TCGA LUAD, LUSC, LIHC, and PAAD datasets for evaluation. iQMAI-derived collagen features were compared to tissue composition, gene expression, and overall survival. ResultsThe iQMAI model shows significant generalization across multiple indications. iQMAI collagen predictions were similar to polarization imaging measurements of the same sample, with a mean structural similarity index (SSIM) of 0.84 (95% CI 0.69-0.93), a mean patch-wise RMSE of 0.04 (95% CI 0.02-0.08), and a linear correlation (R2=0.93). Comparing features of the collagen fibers extracted from iQMAI vs. polarization images yielded similar linear correlations between computed fiber tortuosity, length, width, and relative angle. The relationship between collagen fiber density and fibroblast density was distinct in non-small cell lung cancer (LUAD and LUSC), hepatocellular carcinoma (LIHC), and pancreatic ductal adenocarcinoma (PAAD). In PAAD, fiber density and fiber width were both negatively associated with the LRRC-15 gene expression signature, and increased fiber width was associated with longer overall survival. ConclusionsiQMAI is a deep learning model that accurately predicts collagen from an H&E-stained WSI, allowing for spatially resolved quantification of collagen morphology and enabling investigation of the interplay between collagen and other TME components. We demonstrate an example of the utility of iQMAI-based collagen assessment in PAAD, where collagen features are associated with immunosuppressive cancer-associated fibroblasts and overall survival. Understanding the relationship between collagen, the tumor microenvironment composition, and disease progression may aid the development of effective immunotherapies in PAAD and other cancer types.

pathology↗

bioRxiv2024

Supervised non-negative matrix factorization on cell-free DNA fragmentomic features enhances early cancer detection

BackgroundCell-free circulating DNA (cfDNA) fragments exhibit non-random patterns in their length (FLEN), end-motif (EM), and distance to nucleosome position (ND). While these cfDNA features have shown promise as inputs for machine learning and deep learning models in early cancer detection, most studies utilize them as raw inputs, overlooking the potential benefits of pre-processing to extract cancer-specific features. This study aims to enhance cancer detection accuracy by developing a novel approach to feature extraction from cfDNA fragmentomics. MethodsWe implemented a supervised non-negative matrix factorization (SNMF) algorithm to generate embedding vectors capturing cancer-specific signals within cfDNA fragmentomic features. These embeddings served as input for a machine learning model to classify cancer patients from healthy individuals. ResultsWe validated our framework using two datasets: an in-house cohort of 431 cancer patients and 442 healthy individuals (dataset 1), and a published cohort comprising 90 hepatocellular carcinoma (HCC) patients and 103 individuals with cirrhosis or hepatitis B (dataset 2). In dataset 1, we achieved an AUC of 94% in pan-cancer detection. In dataset 2, our framework achieved an AUC of 100% for HCC vs healthy classification, 99% for HCC vs non-HCC patients classification, and 96% for identifying HCC patients among a mixed group of non-HCC patients and healthy donors. ConclusionThis study demonstrates the efficiency of SNMF-transformed features in improving both pan-cancer detection and specific HCC detection. Our approach offers a significant advancement in leveraging cfDNA fragmentomics for early cancer detection, potentially enhancing diagnostic accuracy in clinical settings.

bioinformatics↗