Biology subjects

Lee, J.

Publications and source records attributed to Lee, J..

At least 109 records · Page 6Linked to original sources

bioRxiv2026

A proteomic atlas of organelle remodeling identifies lysosomal SNX3 as a regulator of Notch signaling in epidermal differentiation

Differentiation of epidermal keratinocytes is accompanied by profound reorganization of intracellular architecture, but how organelle remodeling interfaces with cell fate control is not well understood. Here, we generate a compartment-resolved proteomic map of keratinocyte differentiation and identify extensive remodeling of lysosomes, mitochondria, autophagic vesicles, plasma membrane, and nucleus. Differentiating keratinocytes display coordinated enrichment of lysosomal degradative machinery, vesicular trafficking factors, and mitochondrial metabolic proteins, revealing organelle remodeling as a prominent feature of epidermal differentiation. From the lysosomal proteome, we identify Sorting Nexin 3 (SNX3) as a critical regulator of epidermal homeostasis. SNX3 increasingly localizes to LAMP1-positive vesicles during differentiation, and its loss impairs epidermal differentiation, suppresses Notch signaling, and promotes proliferative gene expression. In vivo, SNX3-deficient skin grafts fail to maintain normal epidermal architecture and instead develop into squamous cell carcinoma. Mechanistically, SNX3 mediates efficient Notch receptor activation, as SNX3 loss reduces nuclear Notch1 and NICD production, whereas NICD re-expression can rescue the differentiation defect. Our study defines a proteomic framework for organelle remodeling during epidermal differentiation and identifies lysosome-associated SNX3 as a key link between endolysosomal trafficking, Notch signaling, and epidermal tissue homeostasis.

cell biology↗

bioRxiv2026

Developing a multi-modal neuroimaging-based BrainAge model across childhood

BrainAge models hold promise as a clinical biomarker for developmental brain health, especially in childhood when there is the potential for early intervention. To distinguish between normative developmental variance and pathological divergence, BrainAge models should reflect the dynamic and diverse neurodevelopmental processes that occur in distinct developmental windows across childhood. We utilized multi-modal neuroimaging data from three pediatric cohorts covering ages 4 to 13 years (n = 1005, 2126 scans), split into Train and Test datasets. Twelve sex-stratified BrainAge models were built stratified by type and different combinations of neuroimaging features. Model types were "Full-Span" models covering the full age range, and "Phase-Specific" models split into early- and late-childhood. We first compared BrainAge estimates in the Test dataset amongst our candidate models, then benchmarked the best-performing model against published pre-trained models and DNA-based biological age measures. Our findings show that a BrainAge model that was phase-specific and consisted of both structural and functional features (cortical thickness, subcortical volumes, and functional network integration measures) showed good prediction of age and best distinguished between healthy and symptomatic subgroups. We present a proof-of-concept for developmental models supporting building BrainAge models of higher temporal resolution that align to different childhood developmental phases.

neuroscience↗

bioRxiv2026

Genomic analysis of BCG unresponsive non-muscle-invasive bladder cancer identifies drivers of sensitivity to intravesical Gemcitabine/Docetaxel

Background and ObjectivesIntravesical gemcitabine/docetaxel (Gem/Doce) is an effective therapy for Bacillus Calmette- Guerin (BCG)-unresponsive non-muscle-invasive bladder cancer (NMIBC), achieving 50% complete responses at 2 years. However, the genomic determinants underlying response and resistance to Gem/Doce remain poorly defined. Our objective was to define the mutational landscape of BCG-unresponsive NMIBC and nominate genomic features associated with response or resistance Gem/Doce. MethodsPatients with BCG-unresponsive NMIBC treated with Gem/Doce were classified as responders (recurrence-free survival [RFS] >12 months) or non-responders (RFS <12 months). Whole-exome sequencing was performed on tumors prior to Gem/Doce treatment (n=23). Single nucleotide variants were identified and annotated using a Cancer Genome Analysis pipeline. Copy number alterations were inferred with ABSOLUTE, and clonal architecture was reconstructed using PhylogicNDT. Key Findings and LimitationsResponders demonstrated significantly prolonged time to high-grade recurrence (3.5 vs 42 months, p<0.001) and cystectomy compared with non-responders (9.5 months vs not reached; p<0.001). Non-responders exhibited higher tumor mutational burden (13.66 vs 8.71; p=0.02) and more frequent whole-genome doubling (2/2 non-responders vs 0/1 responders; p=0.33). Phylogenetic analyses revealed clonal BAP1 and subclonal BRCA2 mutations in responders, whereas non-responders harbored clonal FGFR3 mutations. Limitations include small sample size and retrospective design. Conclusions and Clinical ImplicationsDistinct genomic features underlie differential response to Gem/Doce in BCG-unresponsive NMIBC. In responders, alterations in DNA repair pathways (e.g., BRCA2) may sensitize tumors to chemotherapy, while non-responders with FGFR3 mutations may benefit from alternative targeted strategies. These findings warrant validation in larger cohorts and support the development of biomarker-driven clinical trials. Patient summaryIn this report we analyzed bladder tumors and found that some tumors respond well to treatment because they have defects in repairing DNA, making them more vulnerable to chemotherapy. In contrast, tumors that do not respond to chemotherapy harbor different genetic changes that help them survive and grow. These findings may help physicians choose more effective and personalized treatments in the future.

genomics↗

bioRxiv2026

Determinants of centromeric nucleosome conformation

Centromeric chromatin is defined by the presence of the histone H3 variant CENP-A, which forms a specialized nucleosome required for kinetochore assembly. Compared to canonical H3 nucleosomes, CENP-A nucleosomes exhibit an open DNA conformation that leaves an additional 13 base pairs of DNA accessible at the entry and exit sites. While the CENP-A N-helix has previously been implicated in promoting this enhanced DNA breathing, the contributions of the intrinsically disordered N-terminal tail and adjacent latch regions of CENP-A to nucleosome conformation remain unknown. The intrinsically disordered N-terminal regions of histone H3 are known to facilitate interactions with DNA to stabilize overall nucleosome conformation. Here, we utilized a combination of MNase digestion assays and coarse-grained molecular dynamics simulations of nucleosomes containing targeted swaps of the H3 and CENP-A tail, latch, and N-helix regions to systematically test the contribution of each N-terminal histone region to nucleosomal DNA conformation. Substitution of individual H3 N-terminal regions with those of CENP-A opened the DNA at the entry/exit sites and increased DNA accessibility. While any single CENP-A N-terminal region was sufficient to open the canonical nucleosomal DNA conformation, replacement of any single CENP-A N-terminal region with its H3 counterpart was insufficient to restore the closed DNA conformation characteristic of canonical H3 nucleosomes. Instead, progressive incorporation of multiple H3-derived regions produced increasingly closed DNA conformations, demonstrating that the H3 tail, latch, and N-helices act cooperatively to stabilize wrapped nucleosomal DNA. Taken together, these findings demonstrate that the more restricted DNA breathing of canonical nucleosomes arises from coordinated contributions across multiple N-terminal regions and suggest that the multi-region redundancy in the conformational flexibility of the centromeric nucleosome could emphasize the importance of retaining flexibility in the centromeric nucleosome, even upon post-translational modification and binding to structural proteins.

biophysics↗

bioRxiv2026

Parallels in leg and wing proximal-distal patterning in Holometabola

The wings and legs of insects are both appendages that develop along a proximal-distal (PD) axis and likely share many underlying patterning mechanisms. Comparisons between the two appendages have been detailed in Drosophila melanogaster, a derived insect where both traits develop initially as imaginal discs. Here, we visualise and compare the expression of prominent PD developmental genes in the embryonic legs and larval wings of a lepidopteran species, Bicyclus anynana. We examine the domains of twelve leg gap genes that subdivide legs into distal, medial, or proximal domains, three morphogens, and two genes that refine the PD axis after initial specification. Our results reveal high spatial congruence in the order of PD gene expression between the two appendages. Notably, we observe a distinct loss of the medial domain in the Bicyclus wing compared to the leg, providing evidence for the evolutionary re-patterning of these structures. Comparisons with Drosophila further highlight conserved versus lineage-specific regulatory architectures. These findings suggest a deeply conserved PD patterning logic across Holometabola, while pointing to divergent mechanisms that likely facilitated the morphological innovation of butterfly wings.

developmental biology↗

bioRxiv2026

A self-limiting orexin-habenula circuit for stress resilience

Resilience requires neural systems that mobilise active coping during stress while limiting its persistence to preserve homeostasis under sustained challenge1-4. Here we identify a self-limiting orexin-habenula circuit in which lateral hypothalamic orexin neurons engage aromatic L-amino acid decarboxylase-expressing D-neurons (encoded by Ddc) in the lateral habenula via orexin receptor type 2 (OX2R)5-7. Activation of this pathway increased nucleus accumbens dopamine and promoted active coping and positive valence. Optotagging revealed rapid stress-evoked recruitment followed by post-stress suppression. In lateral habenula neurons, orexin peptides exerted dissociable effects: orexin-A engaged an OX2R-dependent inhibitory programme superimposed on a parallel inward current, whereas orexin-B did not reproduce this inhibitory profile and instead exerted a distinct membrane effect. Chronic stress disrupted this buffering system through coordinated inflammatory activation, promoter methylation and erosion of D-neuron identity and orexin responsiveness. Restoring orexin-A reversed behavioural and molecular deficits through OX2R-dependent suppression of nuclear factor kappa B signalling, preservation of Tet2 expression, and demethylation-linked maintenance of the Ddc programme. Together, these findings define a self-limiting orexin-habenula resilience circuit that enables adaptive coping while constraining stress-induced vulnerability.

neuroscience↗

bioRxiv2026

Label-free quantitative 3D mapping of collagen architecture by holotomography

Fibrillar collagen architecture regulates tissue mechanics and cell-matrix interactions, but following how collagen-rich matrices assemble and remodel in three dimensions requires non-destructive measurements that can link structure to matrix-associated mass without exogenous labels. Here we present holotomography (HT) as a label-free platform for volumetric imaging and quantitative profiling of reconstituted fibrillar collagen. Volumetric RI tomograms resolved fibrillar collagen networks without exogenous labels, and co-registration with second-harmonic generation (SHG) microscopy supported the correspondence between HT-resolved RI structures and SHG-positive fibrillar collagen. Curvelet-transform-based segmentation of representative axial sections converted the tomograms into fiber-level descriptors, including morphology, orientation and RI-derived fiber-associated dry mass. At the matched 0.8 mg/mL condition, HT resolved distinct network organization, with type I forming thicker, more sparsely distributed fibers and type III a finer, denser network, while mean fiber-associated RI remained comparable between subtypes. Across the concentration series, type I showed stronger increases in fiber width and RI-derived mass, whereas type III remained comparatively fine. Time-lapse HT further captured RI dynamics during type I fibrillogenesis and visualized local cell-matrix reorganization in HT1080 cell-embedded gels, capturing cells and surrounding matrix in a single label-free measurement. Together, these results position HT as a complementary label-free approach for volumetric visualization and RI-based quantification of collagen-rich extracellular matrices.

cell biology↗

bioRxiv2026

From Garden Soil to Bio-electricity: Utilizing Rhizosphere Diversity for Microbial Fuel Cell Operation

This study investigates the potential of the garden rhizosphere as a source of electrochemically active bacteria (EAB) for operating microbial fuel cells (MFCs). We evaluated a diverse array of garden flora, including vegetables (lettuce, Chinese cabbage), flowering plants (August lily, peppermint), and woody species (pine, oak, ginkgo, and bush clover). Among the tested groups, MFCs inoculated with peppermint and ginkgo rhizosphere microbiotas exhibited the highest current densities within their respective categories, significantly outperforming control groups without plant components. 16S rRNA gene microbial community analysis revealed that the initial rhizosphere environment acts as a decisive selective pressure, shaping distinct anode biofilms based on plant types (herbaceous vs. woody). Despite these structural differences in microbial assembly, high current generation was achieved in both peppermint and ginkgo systems, suggesting a high degree of functional redundancy within the rhizosphere-derived consortia. These findings demonstrate that various garden ecosystems can serve as robust biological reservoirs for MFC operation, where diverse microbial configurations are capable of sustaining efficient bio-electrochemical energy conversion.

microbiology↗

bioRxiv2026

Metabolic signals regulate resuscitation speed of antibiotic persister bacteria during infection

All living organisms adjust their metabolism in response to environmental changes. Under unfavorable conditions, organisms enter a state of dormancy by halting metabolism, enabling survival. Dormant bacteria become highly tolerant to antibiotics-a phenomenon called persistence. Here, we demonstrate that selective metabolic reprogramming controls the resuscitation speed of persister after antibiotic exposure. Using multi-omics and in silico modeling, we found that dormant bacteria reprogram metabolic pathways to modulate persister awakening. Accumulation of L-serine and reduction of arginine drive rapid resuscitation. L-serine promotes cysteine biosynthesis and motility while reducing energy metabolism to facilitate rapid resuscitation. In contrast, arginine slows regrowth from dormancy by enhancing ethanol-aldehyde and energy metabolism. L-serine and arginine can, respectively, promote or inhibit the regrowth of antibiotic persister cells in macrophages and mouse models, and regulate the awakening speed of Salmonella, E. coli, and methicillin-resistant Staphylococcus aureus (MRSA). These findings suggest new strategies to target chronic bacterial infections. TeaserL-serine speeds and arginine slows the awakening of antibiotic persisters, revealing targets for chronic infection.

microbiology↗

bioRxiv2026

Bipartite DNA binding domain of transcription factor BCL11B binds clustered short DNA sequence motifs

B-cell leukemia/lymphoma 11B (BCL11B), despite its name, is a key regulator of T-cell development, specification, and T-cell malignancies. BCL11B contains a bipartite DNA binding domain composed of two C2H2 zinc finger arrays: low-affinity ZF2-3 and high affinity ZF4-6. These arrays function as homotypic modules that recognize similar six-nucleotide motifs, TG(O_SCPLOWNC_SCPLOW)CC(O_SCPLOWCC_SCPLOWO_SCPCAP/C_SCPCAPO_SCPLOWTC_SCPLOWO_SCPCAP/C_SCPCAPO_SCPLOWAC_SCPLOW), as seven of the eight DNA base-contacting residues are conserved between them. The most conserved interactions involve GG dinucleotides, contacted by arginine and lysine residues at key base-interacting positions in ZF3 and ZF5. The two ZF arrays are connected by a long [~]300-residue linker that provides flexibility in how the arrays engage DNA, allowing ZF2-3 and ZF4-6 binding to the same or opposite strands with variable orientation, spacing and positioning along the DNA. This extended linker is enriched in serine/threonine, acidic residues (aspartate/glutamate), and structural residues (glycine/proline), providing additional layers of transcriptional regulation possibly through post-translational modification, electrostatic modulation, and/or condensate formation. We also examined six missense mutations in base-interacting residues, that are associated with neurodevelopmental disorders. Substitutions replacing bulky, positively charged arginine or lysine with smaller or hydrophobic residues likely reduce DNA-binding affinity and/or specificity, whereas substitutions between asparagine and lysine may alter base recognition preferences.

biochemistry↗

bioRxiv2026

G2-to-G0 cell cycle exit underlies sensitivity to ATR inhibition via the p53-p21-RB1 axis

Ataxia-telangiectasia and Rad3-related (ATR) is an essential DNA damage response kinase that protects genome integrity by controlling cell cycle checkpoints, regulating origin firing, stabilizing replication forks, and signaling DNA repair. Due to hyper-proliferation, cancer cells depend on ATR for survival, implicating ATR inhibitors as promising therapeutics. However, variable tumor responses to ATR inhibitors highlights the need to uncover the determinants of cell fate. Here, we show breast cancer sensitivity to ATR inhibition correlates with the appearance of pan-nuclear DNA damage. The fate of these cells is driven by a p53-p21-RB1 axis that triggers a G2-to-G0-like cell cycle exit and is buffered by the p53 inhibitor MDM2. MDM2 inhibition lowers the DNA damage threshold for cell cycle exit and robustly targets ATR inhibitor-resistant cells. Our work reveals cell cycle plasticity as a mechanism determining cell fate during ATR inhibition and identifies MDM2 as a target for increasing ATR inhibitor efficacy.

molecular biology↗

bioRxiv2026

Tuned inhibitory control of neuronal firing thresholds explains predictive sensorimotor behavior

Prior expectations guide sensorimotor behavior when sensory information is uncertain, yet the cellular mechanisms underlying this integration remain elusive. Here, we investigate how priors in sensory motion direction shape neural population dynamics by combining recurrent neural network (RNN) modeling with macaque smooth pursuit behavior and middle temporal area (area MT) electrophysiology. Our RNN model reveals that prior expectations are implemented by elevating firing thresholds in neurons tuned away from the expected direction. This selective inhibition sharpens population tuning and reduces behavioral variability under weak sensory information conditions. We validated this prediction in vivo: delta-band local field potentials in area MT, where the phase reflects neural excitability states, exhibited direction-specific phase shifts that scaled with the deviation from the expected direction. These findings demonstrate that prior expectations enhance behavioral reliability through tuned inhibitory control of neuronal excitability, providing a mechanistic link between Bayesian inference and cortical circuit dynamics.

neuroscience↗

bioRxiv2026

Lactylation landscape of mitochondrial proteins in myocardial infarction

Metabolic reprogramming is a hallmark of myocardial infarction (MI), in which cardiomyocytes shift from fatty acid oxidation to anaerobic glycolysis, leading to elevated lactate production and mitochondrial dysfunction. Lactylation, a recently described lysine post-translational modification, has emerged as a metabolic signaling mechanism; however, its role within mitochondria during MI remains poorly understood. Here, we define the mitochondrial lactylome following MI and examine how modulation of lactate transport influences mitochondrial metabolism and redox homeostasis. Using quantitative proteomics, we identify extensive remodeling of mitochondrial protein lactylation after MI, affecting enzymes involved in bioenergetics, redox regulation, and metabolic control. Pharmacological inhibition of monocarboxylate transporter-1 (MCT1) using AZD3965 further reshapes the mitochondrial lactylome, increasing lactylation of specific metabolic and redox-associated proteins without uniformly exacerbating mitochondrial dysfunction. Despite sustained impairment of global cardiac function, MCT1 inhibition attenuates post-MI fibrosis and inflammation and partially restores mitochondrial respiratory capacity. Consistent with in vivo findings, genetic or pharmacological inhibition of MCT1 in hypoxic cardiomyocytes-derived cells reduces mitochondrial reactive oxygen species, decreases inhibitory pyruvate dehydrogenase phosphorylation, and improves mitochondrial bioenergetics. Together, these findings reveal that mitochondrial lactylation is a context-dependent regulator of mitochondrial metabolism and redox balance following MI. Rather than acting solely as a pathological modification, lactylation integrates lactate availability with mitochondrial function to influence inflammatory and fibrotic remodeling, highlighting mitochondrial metabolic plasticity as a potential therapeutic target in ischemic heart disease. HighlightsO_LIMyocardial infarction (MI) increases mitochondrial protein lactylation, with 361 identified lactylated proteins. C_LIO_LIAZD3965-mediated MCT1 inhibition further elevates mitochondrial lactylation. C_LIO_LIDistinct alterations in mitochondrial proteins and pathways (TCA cycle, amino acid metabolism, gene expression) were observed. C_LIO_LIAZD3965 reduces cardiac fibrosis and inflammation and partly improves mitochondrial respiration post-MI, but cardiac function remains impaired. C_LI O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=115 SRC="FIGDIR/small/718938v1_ufig1.gif" ALT="Figure 1"> View larger version (47K): org.highwire.dtl.DTLVardef@b5a7b3org.highwire.dtl.DTLVardef@14ea92org.highwire.dtl.DTLVardef@1343a29org.highwire.dtl.DTLVardef@1d67716_HPS_FORMAT_FIGEXP M_FIG Graphical Abstract C_FIG

cell biology↗

bioRxiv2026

Reciprocal repulsions enforce heterotypic dendrite segregation in an olfactory circuit

How dendrites of different neurons segregate into discrete spatial domains during neural circuit assembly is poorly understood. Here, using the Drosophila olfactory system, we found that heterophilic interactions between two cell-surface proteins Teneurin-m (Ten-m) and Capricious (Caps) drive dendrite segregation. Ten-m and Caps are expressed in largely inverse patterns across projection neuron (PN) types when PNs are establishing their dendritic territories. Loss of Ten-m in Ten-m+ PNs causes their dendrites to invade Caps+ territories, whereas loss of Caps in Caps+ PNs causes dendrite invasion into Ten-m+ territories. Structure-guided mutations that abolish Ten- m-Caps binding disrupt dendrite segregation, whereas the same mutation on Ten-m preserves its homophilic attraction in a synaptic partner matching assay. These results support a model in which mutual repulsions between two inversely expressed cell-surface proteins drive dendrite segregation into discrete glomerular territories.

neuroscience↗

bioRxiv2026

Eggerthella lenta evades bacteriophage through reversible megabase-scale inversions of capsular polysaccharide gene clusters

Bacteriophages are a promising tool for microbiome editing, yet their development has been constrained by limited insights into bacteriophage-host interactions within their shared mammalian body habitat. We isolated a lytic phage {Phi}KL11 that efficiently targets a disease-associated member of the human gut microbiota, Eggerthella lenta, during in vitro growth. However, {Phi}KL11 selects for a pre-existing and reversible bacteriophage-resistant sub-population in mice. Long-read sequencing revealed a massive genomic inversion event, representing >50% of the E. lenta genome, enriched in response to bacteriophage infection. Transcriptomics linked this inversion to the altered expression of three capsular polysaccharide synthesis (CPS) gene clusters and transmission electron microscopy confirmed differential capsule production. Finally, we show that {Phi}KL11 has a broad host range attributable to CPS and other strain-variable genes. These findings suggest a previously unrecognized strategy for phage evasion in the gut, involving megabase-scale genomic inversions and reversible capsule variation driving phage resistance.

microbiology↗

bioRxiv2026

Comparative Analysis of Relative Ligand Binding Free Energy Simulation Methods: Amber-TI, GROMACS-NETI, OpenMM-FEP, and BLaDE-MSLD

Structure-based drug design has become increasingly important in the pharmaceutical industry for accelerating the discovery of effective drug candidates. In particular, ligand binding free energy serves as a critical metric for predicting drug efficacy during the key stages of hit discovery and lead optimization. Continuous progresses have been made in the prediction of ligand binding free energies, but direct comparisons of different methods using the same force field remain challenging due to their unique implementations into different simulation engines. In this study, we present a direct comparison of four popular methodologies (Amber-TI, GROMACS-NETI, OpenMM-FEP, and BLaDE-MSLD) for calculating relative binding free energies ({Delta}{Delta}Gbind) with the same Amber protein and ligand force fields using MolCube Alchemical Free Energy Simulator (MolCube-AFES), which provides an input generation workflow to support {Delta}{Delta}Gbind calculations of all four methods. We used 80 alchemical transformations (among the JACS benchmark set by Wang et al.) and two additional applications to compare the predicted {Delta}{Delta}Gbind from the four methods against experimental measurements. All four methods reproduced experimentally observed trends with most transformations within {+/-}2 kcal/mol from experiments and show broadly comparable accuracy with no statistically significant performance differences across the benchmark dataset. These results demonstrate that MolCube-AFES enables controlled, cross platform benchmarking and show that all four different alchemical free energy methods deliver statistically equivalent accuracy, with method selection guided by workflow requirements such as throughput, portability, and perturbation network design rather than expected differences in performances.

biophysics↗

bioRxiv2026

Landscape-scale navigation unlocks antibody CDR structural logic for AI-guided rescue and therapeutic optimization

While AI offers transformative potential for therapeutic antibody design, the lack of ground-truth data fundamentally constrains our ability to model the epistatic topology of fitness landscapes. Here, we establish a high-throughput workflow to characterize tens of thousands of antibody variants per week with gold-standard biophysical precision. By combinatorially assembling functional variants from deep mutational scanning, we charted antibody fitness landscapes comprising over 17,000 data points, which revealed an extremely rugged, non-navigable epistatic topology. Yet, navigating at this unprecedented scale enabled the discovery of rare peak clusters exhibiting simultaneous enhancements in affinity and productivity. Strikingly, ProteinMPNN predicted the CDR-dependent productivity landscape with remarkable accuracy, suggesting that sequence-structure compatibility within CDRs gates cellular productivity. This insight enabled a structure-guided rescue strategy combining AlphaFold3 and ProteinMPNN, which successfully restored the cellular productivity of high-affinity, low-productivity clones via single amino acid substitutions. Two elite variants drawn directly from peak clusters further demonstrated 20- to 100-fold in vivo efficacy gains in a murine psoriasis model. Our findings establish CDR structural fitness as a fundamental determinant of antibody cellular productivity and validate landscape-scale navigation as a powerful framework for therapeutic antibody optimization.

bioengineering↗

bioRxiv2026

Machine Learning Approach for Enumeration of Circulating Cells with Diffuse in vivo Flow Cytometry

SignificanceDiffuse in vivo flow cytometry (DiFC) is an emerging technique for enumerating rare, fluorescentlylabeled circulating tumor cells (CTCs) in small animals without drawing blood samples. DiFC uses detection of transient fluorescent peaks in time-series data. Previously, we used a simple amplitude threshold-based method for identifying peak candidates, but it ignores potentially useful information in peak shape that could reduce false-positive detections from instrument noise and increase detection efficiency of lower-amplitude peaks. AimTo develop a machine learning (ML)-integrated signal processing approach for improved CTC enumeration using DiFC by distinguishing CTC peaks from artifacts. ApproachWe developed an ML-integrated approach that incorporates a convolutional neural network (CNN) classifier. The CNN was trained to distinguish CTC peaks from artifacts by analyzing peak amplitude and temporal shape characteristics. Performance was validated on in-silico, control, and CTC-bearing mouse datasets. ResultsThe CNN classifier achieved accuracy, precision, sensitivity, and specificity exceeding 98% on test data. Compared with our previously published threshold-based approach, the ML-integrated method increased the number of correctly identified CTCs and their flow direction while reducing false detections across validation datasets. ConclusionsThe ML-integrated approach significantly improves DiFC CTC enumeration, enabling robustness against artifacts in noisy conditions.

bioengineering↗