Biology subjects

Zhao, X.

Publications and source records attributed to Zhao, X..

At least 55 records · Page 3Linked to original sources

bioRxiv2026

RISOP, a Reference-Assisted Approach for Enhanced Identification of Oxidized Phospholipids

Oxidized phospholipids (OxPLs) play critical roles in inflammation, ferroptosis, and other oxidative stress-associated processes, yet their systematic characterization in biological systems remains a major analytical challenge owing to their low abundance and vast structural diversity. Here we report RISOP (Reference-Assisted Identification of Sample-specific Oxidized Phospholipids), an untargeted LC-MS/MS workflow that leverages chemically enriched OxPL reference pools to substantially improve OxPL annotation. Reference pools encompassing diverse oxidative modifications and a wide abundance range were experimentally generated using Fenton reaction and H2O2 treatment, providing broad coverage of OxPLs. Reference-assisted integrative analysis yielded a more than twofold increase in identified OxPL species in biological samples under elevated oxidative stress, as demonstrated in ML210-treated cells, a model of ferroptotic stress. We further show that the widely used BODIPY C11 lipid peroxidation probe captures cellular oxidative burden in only a subset of OxPLs identified by RISOP, highlighting the importance of untargeted, comprehensive OxPL profiling. Overall, RISOP provides a versatile platform readily applicable to other classes of oxidized complex lipids for comprehensive characterization in physiological and pathological contexts. Abstract Graphic O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=105 SRC="FIGDIR/small/728773v1_ufig1.gif" ALT="Figure 1"> View larger version (24K): org.highwire.dtl.DTLVardef@482111org.highwire.dtl.DTLVardef@173126corg.highwire.dtl.DTLVardef@1be7238org.highwire.dtl.DTLVardef@3cf7f2_HPS_FORMAT_FIGEXP M_FIG C_FIG

systems biology↗

bioRxiv2026

Repetition Selectively Reconfigures Offset-Defined Temporal Integration in Auditory Cortex

Auditory perception requires integrating sound over time, but whether cortical integration windows are fixed or adaptively shaped by recent sensory history remains unclear. Because auditory offset responses are shaped by the temporal structure preceding sound termination, they provide a readout of offset-defined temporal integration. Here, we recorded single-unit activity in awake rat auditory cortex while presenting 512-ms click trains with inter-click intervals of 1, 2, 4, 8 or 16 ms. Each stimulus was repeated ten times, allowing repetition number to index increasing adaptation. Across 271 offset-responsive neurons, repetition strongly compressed the maximal ICI capable of evoking a significant offset response and produced a smaller population-level shift in preferred ICI. These preferred-ICI changes were heterogeneous across neurons, with bidirectional shifts at the single-neuron level. In contrast, the effective temporal window measured with irregular-regular click trains remained stable at the population level, despite neuron-level variability. AAF and A1 further showed distinct repetition-dependent tuning profiles. These findings suggest that adaptation selectively reconfigures ICI-based offset tuning while preserving a population-level local regularity window, revealing flexible and stable components of cortical temporal integration.

neuroscience↗

bioRxiv2026

In-cell structures visualize human pre-ribosome assembly in the nucleolus

Ribosome biogenesis is an essential multi-step process. In eukaryotes, it is orchestrated within the nucleolus, a hallmark multi-layered compartment of the nucleus. Structures of pre-ribosomes have been characterized in vitro, but their assembly pathways within human nucleoli have not been described. Here, we used cryogenic correlative light and electron tomography to visualize the molecular landscapes within HeLa cell nucleoli and obtained in-cell structural snapshots of both ribosomal subunit precursors, the SSU processome and the pre-60S. These recapitulate major states resolved previously in vitro while additionally revealing critical interaction partners, including the RNA exosome, rixosome, and nuclear export receptor CRM1-RanGTP. We further show how pre-ribosome assembly is altered upon RNA polymerase I inhibition. Our study combines molecular structures with cellular context to elucidate the spatiotemporal assembly pathways and regulation of human ribosome biogenesis.

cell biology↗

bioRxiv2026

Graph-based pangenome provides insights into the adaptive evolution of Cucurbita pepo

Understanding how crops respond to environmental variation is crucial for biodiversity conservation and food security. Cucurbita pepo (pumpkin, squash, gourd) is one of the first domesticated crop species and exhibits remarkable phenotypic and ecological diversity, making it a powerful system for investigating the genomic basis of adaptation. Here, we constructed a graph-based C. pepo pangenome using nine chromosome-level assemblies and identified 229,431 high-confidence structural variants (SVs) that were genotyped across 206 wild and cultivated accessions. Our results demonstrate that C. pepo underwent parallel domestication, with two deeply diverged gene pools independently giving rise to the pepo and ovifera cultivated lineages, followed by expansion into diverse environments that produced strong signatures of differentiation largely mediated by young adaptive alleles. Single-nucleotide polymorphisms (SNPs) and SVs contribute complementary dimensions of environmental responsiveness. Biogeographical modeling predicts continued range contraction of wild relatives and elevated genetic offset in Eastern North American populations under projected future climates. Populations with higher genetic load harbor fewer adaptive variants and exhibit greater predicted maladaptation, indicating that deleterious mutations constrain adaptive potential. These findings highlight how genomic variants, adaptive diversity, and genetic load together shape environmental adaptation and inform conservation and crop improvement.

genomics↗

bioRxiv2026

EEG-Neurofeedback Targeting Gamma Oscillations at the Parieto-Occipital Region Reduces Pain Perception

Pain is closely associated with both spontaneous and stimulus-evoked gamma oscillations, which differ in their phenomenology and their relationship to pain perception. These associations make gamma oscillations a promising target for pain interventions, such as closed-loop real-time neurofeedback (NFB). While modulating stimulus-evoked gamma oscillations is technically challenging due to their concurrent occurrence with pain perception, targeting spontaneous gamma oscillations is both feasible and practical, as it could influence subsequent cortical nociceptive processing and reduce pain. To test this mechanistic hypothesis, we developed a novel NFB training protocol aimed at increasing spontaneous gamma oscillations in the parieto-occipital region to alleviate pain. Eight-eight healthy, right-handed subjects were randomly assigned to either an active or sham NFB group and completed three sessions of NFB training. During each session, subjects underwent EEG recording and received randomized laser stimulation while watching a video. After three training sessions, approximately 52% of subjects in the active group successfully increased the spectral power of spontaneous gamma oscillations in the parieto-occipital region. Crucially, the spectral power of spontaneous gamma oscillations was significantly and negatively correlated with subjective pain intensity after the third NFB session. Additionally, subjects who responded successfully to active NFB showed significant reductions in pain intensity, unpleasantness, and laser-evoked potentials compared to the sham group. Together, these findings establish a causal relationship between spontaneous gamma oscillations in the parieto-occipital region and pain perception, potentially offering a novel NFB-based therapeutic strategy for pain management.

neuroscience↗

bioRxiv2026

DOC Release and Physiological Response of Zostera marina Under Light and Nutrient Gradients

The seagrass bed, known as an important blue carbon ecosystem, not only has robust carbon sequestration capacity, but also releases the fixed organic carbon into the water column in the form of dissolved organic carbon (DOC). Zostera marina is the main constructive species of the seagrass meadows of northern China, but the regulatory mechanism of its DOC release in response to environmental changes still remains unclear. Adult Z. marina was used as the experimental material in this research to investigate the effects of light intensity and nutrient concentration on its DOC release, and to observe its physiological responses. By using a total organic carbon (TOC) analyzer, dissolved oxygen monitoring, transmission electron microscopy (TEM), and transcriptome analysis, we explored how light and nutrients affect DOC release and the underlying physiological mechanisms in Z. marina. The results showed that both elevated light intensity and short-term nutrient enrichment promoted organic carbon release from Z. marina to the environment. Increased light enhanced the net primary productivity (NPP) of Z. marina, while the magnitude of DOC release increased more than NPP itself. Nutrient enrichment reduced NPP while raising the proportion of DOC released by Z. marina. Transcriptome analysis showed that nutrient enrichment significantly affected pathways, including flavonoid biosynthesis and fatty acid degradation. This study demonstrated that changes in light and nutrients can significantly influence DOC release and carbon allocation patterns in Z. marina, providing a theoretical basis for blue carbon accounting and ecological restoration evaluation in seagrass beds.

plant biology↗

bioRxiv2026

In situ molecular architecture of the mammalian sperm nuclear vacuole

Direct identification of macromolecular complexes in their native context remains a major barrier to unbiased biological discovery. This challenge is particularly acute in mammalian sperm nuclei, in which condensed chromatin is interspersed with poorly understood phase-separated compartments termed nuclear vacuoles. Vacuoles are associated with reduced fertilization efficiency, yet their composition remains unclear. Here we combine high-resolution in situ cryo-electron tomography (cryo-ET) with AlphaFold docking to identify vacuole components as proteasomes, the proteasome activator PA200, and ferritin. In situ structures at resolutions up to 3.8 [A] reveal distinct proteasome-PA200 associations and gating states, consistent with a stepwise activation mechanism. Ferritin assemblies exhibit heterogeneous mineralization states and directly contact chromatin. Together, these findings establish the molecular organization of sperm nuclear vacuoles and implicate protein turnover and metal homeostasis in shaping the nuclear landscape, while demonstrating the power of in situ cryo-ET to resolve protein identity and conformational dynamics in native cellular environments.

biochemistry↗

bioRxiv2026

Replicability of unsupervised deep learning derived image phenotypes

Unsupervised deep-learning image phenotypes derived from brain MRI are propelling imaging genetics to link brain structure to genetic variation. However, their replicability across data sets has not been sufficiently evaluated, raising questions about whether they capture robust biological structure or reflect training-specific artifacts. Here, we assess the replicability of unsupervised deep-learning image phenotypes under variation in model initialization, data partitioning, and cohort, directly evaluating their stability across experimental conditions. We trained multiple models under (i) different training batch random seeds, (ii) cross-validation splits, and (iii) independent datasets (UKB and ADNI), across CNN and ViT architectures. We then derived representations from a separate UKB discovery cohort (N = 22,985) for both trained models and random initialized models without training. The representation stability was assessed using centered kernel alignment (CKA; mean ViT 0.74 vs random 0.27) and kernel canonical correlation analysis (KCCA; mean ViT 0.84 vs random 0.60), as well as genetic discovery stability using loci overlap ratio (mean ViT 0.45 vs random 0.08). We further applied weighted MAXVAR generalized CCA to 12 embeddings to extract a shared 30-dimensional subspace. Our result showed that UDIPs exhibit statistically significant stability (CKA, KCCA t test p < 0.001) across training perturbations and preserve biologically meaningful structure (loci overlap ratio t test p <0.001) across cohorts, supporting their use in imaging genetics.

bioinformatics↗

bioRxiv2026

Microplastics inhibit macrophage bioenergetics impairing homeostatic function and immune responsiveness

Since the 1950s, micro- and nanoplastics (MNPs) have become omnipresent, representing a novel environmental hazard which continually deposits in our airways. Pulmonary macrophages (pMacs) orchestrate the balance between inflammation and tolerance required for homeostasis of the lung and are among the first immune cells to encounter inhaled MNPs. Yet, how pMacs react to plastic deposition in the lung and implications for disease remain unknown. Here, we exposed mice in vivo, human precision-cut lung slices (hPCLS) ex vivo, and monocyte-derived macrophages and cell lines to polystyrene MNPs in vitro. MNP deposition in the lung and extrapulmonary tissues was determined over a 1-week period and pMacs from MNP-laden lungs isolated for RNA-sequencing. We compared the effects of MNPs or diesel exhaust particulate exposures on hPCLS viability and metabolism, monocyte-derived macrophage transcription, and macrophage mitochondrial function, inflammation, and antigen presentation. MNPs readily translocated the lung and were observed in all organs examined within 1-day. pMacs from MNP-exposed mice expressed transcriptional pathways associated with endocrine system disorders, tissue remodeling, and malignant disease. Macrophage phagocytosis was impaired through decreased mitochondrial function which could be rescued pharmacologically. MNPs inhibited the ability of macrophages to effectively present OVA-antigen preventing TCR-specific activation, an effect that could be restored by blocking PD-1/PD-L1. These findings indicate that MNPs impair macrophages via unique mechanisms linking phagocytic and bioenergetic dysfunction. Loss of antigen-presenting capabilities in MNP-laden macrophages may compromise immunosurveillance. As such, MNPs have the potential to increase susceptibility to lung disease independent of the conventional mechanisms of inflammation and oxidative stress. Clinical relevanceO_LIBioaccumulation of micro- and nanoplastics in macrophages impairs their ability to function as antigen-presenting cells increasing susceptibility to pathogenic and malignant disease. C_LIO_LIPulmonary macrophages residing in micro- and nanoplastic laden lungs possess transcriptional profiles associated with endocrine system disorders, gastrointestinal disease, and cancers. C_LI

immunology↗

bioRxiv2026

CAD rewires nucleotide metabolism to drive oxidative stress in myocardial ischemia/reperfusion injury

Myocardial ischemia/reperfusion (I/R) injury is a major determinant of infarct size and clinical outcome, yet effective therapies remain limited. Although metabolic remodeling is central to I/R pathology, the contribution of nucleotide biosynthesis remains unclear. Here, we identify CAD, the multifunctional rate-limiting enzyme of de novo pyrimidine biosynthesis, as a previously unrecognized regulator of myocardial reperfusion injury. CAD activation exacerbated cardiomyocyte death during simulated I/R, whereas CAD knockdown or pharmacological inhibition was protective in vitro. Mechanistically, CAD enhanced dihydroorotate dehydrogenase (DHODH)-dependent electron transfer, increased the CoQH2/CoQ ratio, and promoted complex I reverse electron transport (RET), thereby amplifying mitochondrial ROS. In parallel, CAD suppressed de novo purine synthesis, causing purine insufficiency, DIS3L-dependent RNA decay, and cytosolic ROS. Importantly, cardiomyocyte-specific CAD deletion protected against cardiac I/R injury in vivo. Together, these findings establish CAD as a metabolic hub linking nucleotide flux to dual-compartment ROS signaling and identify nucleotide metabolism as a therapeutic vulnerability in myocardial I/R injury. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=187 HEIGHT=200 SRC="FIGDIR/small/725578v1_ufig1.gif" ALT="Figure 1"> View larger version (87K): org.highwire.dtl.DTLVardef@14c4fedorg.highwire.dtl.DTLVardef@1138ed2org.highwire.dtl.DTLVardef@1057d8borg.highwire.dtl.DTLVardef@17546bd_HPS_FORMAT_FIGEXP M_FIG C_FIG

cell biology↗

bioRxiv2026

Targeting cancer-associated cell surface RNAs with oligonucleotide-drug conjugates enables broad antitumor activity

Recent studies have revealed the presence of RNAs on mammalian cell surfaces, yet the linkage of cell surface RNAs (csRNAs) to cellular states and their potential as extracellular accessible drug targets remains unexplored. Here we develop Cell-surface and Intracellular RNAs Co-mapping (CIRCmap), a highly multiplexed in situ profiling approach that simultaneously detects thousands of cell-surface and intracellular RNAs at single-cell resolution. Using CIRCmap, we uncover that csRNA distributions are correlated with cellular states and identified cancer-associated csRNAs. Integrative analysis of cell-surface and intracellular RNAs within the same cells implies that the csRNAs undergo endocytosis and endolysosomal trafficking, which is further supported by perturbation experiments and co-localization visualization. We design oligonucleotide-drug conjugates (ODCs) that target cancer-associated csRNAs for endocytic delivery of cytotoxic payloads selectively to cancer cells. ODCs exhibit broad antitumor activity in cell lines and an in vivo mouse tumor model, opening a promising new avenue for targeted cancer therapy.

cell biology↗

bioRxiv2026

Helminth-remodeled microbial indole-3-lactic acid drives AhR-dependent disease tolerance

Helminths systemically suppress host immunity, yet whether they impose immune tolerance by rewiring host-associated microbial metabolism remains unclear. Here we show that Trichinella spiralis infection remodels intestinal tryptophan metabolism to generate an AhR-dependent regulatory immune state. T. spiralis infection enriched the commensal bacterium Ligilactobacillus murinus, which converted tryptophan into indole-3-lactic acid (ILA), a microbial metabolite that directly engaged the aryl hydrocarbon receptor. Antibiotic-mediated microbiota depletion abolished infection-induced ILA accumulation, AhR activation and Treg/Th17 rebalancing, whereas fecal microbiota transplantation from infected donors or supplementation with L. murinus or ILA restored these effects. Pharmacological blockade or genetic deletion of AhR eliminated the ability of T. spiralis, L. murinus and ILA to restrain LPS-induced cytokine-storm-like lung inflammation, establishing AhR as an essential host node in this circuit. Extending these findings to viral inflammatory disease, oral ILA improved survival and reduced pulmonary immunopathology in SARS-CoV-2-infected K18-hACE2 mice. Re-analysis of human COVID-19 metabolomic data further revealed reduced circulating ILA in severe disease. These findings define a helminth-remodeled microbial tryptophan metabolic pathway that promotes disease tolerance and identify the ILA-AhR axis as a candidate postbiotic strategy for limiting hyperinflammatory tissue injury.

immunology↗

bioRxiv2026

Modeling human B cell development with pluripotent stem cells

The ability to generate functional B cells from human pluripotent stem cells (hPSCs) would open new opportunities to develop novel B cell-based therapies to treat a range of human diseases and disorders. Towards this goal, we established a protocol that promotes the efficient development of B lineage cells from definitive hematopoietic progenitors generated from different hPSC lines. Flow cytometric and multi-omic scRNA-seq analyses revealed that B cell development from hPSCs transitions through the well-established pro-B, pre-B and naive B cell stages, accurately recapitulating B lymphopoiesis in the human adult bone marrow. Importantly, the naive B cells generated with this approach could be induced to mature into plasma cells that secrete antibodies and undergo class switching. Analyses of signaling pathways that regulate B lymphopoiesis in these cultures uncovered a potent inhibitory effect of IL-7 on functional IgH rearrangement, resulting in the development of abnormal cells that failed to undergo pre-B cell maturation. Finally, analysis of the different hPSC-derived hematopoietic programs revealed that both definitive and yolk sac progenitors display B cell potential, indicating that there are distinct developmental sources of human B lineage cells. Taken together, these findings demonstrate the efficient generation of B cells from hPSCs and, in doing so, provide a system for further investigating the earliest stages of human B lymphopoiesis and a source of appropriately staged plasma cells for future therapeutic applications.

developmental biology↗

bioRxiv2026

IL-36γ Drives Th9 cell Differentiation via IκBζ to Sustain Antitumor Immunity

Th9/Tc9 cells polarized with TGF-{beta} and IL-4 exhibit superior antitumor efficacy, prompting extensive efforts to optimize their differentiation protocols and augment therapeutic potency. In the current study, we identified IL-36{gamma} as a potent cytokine that synergizes with TGF-{beta} to robustly drive Th9 differentiation both in the presence and absence of IL-4. IL-36{gamma}-programmed Th9 cell subsets exhibited phenotypic and transcriptional profiles identical to classic Th9 cells. Mechanistically, IL-36{gamma} drove Th9 cell differentiation through amplifying key signaling pathways such as STAT6, STAT5, and NF-{kappa}B that are essential for classic Th9 programming. Notably, we uncovered a novel regulatory axis wherein IL-36{gamma} upregulates the transactivation factor I{kappa}B{zeta}, which directly governs Th9 lineage specification. In in adoptive cell therapy (ACT) models, IL-36{gamma}-polarized Th9 cell subsets demonstrated enhanced antitumor efficacy, attributable to their sustained persistence, reduced exhaustion markers and stem-like/memory properties. Collectively, this study elucidates a previously unrecognized I{kappa}B{zeta}-dependent mechanism underpinning Th9 differentiation and highlights the translational potential of IL-36{gamma}-engineered Th9 cells as a valuable ACT strategy for refractory tumors.

immunology↗

bioRxiv2026

Replication fork remodeling proteins, Smc5/6 and Rtt107, promote palindrome-mediated genome instability

Palindromic sequences are a potent source of genomic instability that can lead to cancer and hereditary diseases in humans. Molecular evidence shows that palindrome instability results from the formation of secondary structures, such as hairpins and cruciforms, which are cleaved by structure-specific nucleases. However, the mechanisms underlying cruciform formation and cleavage at palindromic sequences in eukaryotic cells remain incompletely understood. Here, we describe a pathway for secondary structure formation and chromosomal breakage at palindromes involving DNA helicase Mph1, Rad51 recombinase, Rad54 ATPase DNA strand remodeler, Rtt107 scaffold protein, and the multifunctional Smc5/6 complex. Deletion or mutation of any of these components results in a similar reduction in double-strand breaks at an Alu palindrome and a significant decrease in chromosomal rearrangements. We propose that Mph1, Rad51, and Rad54 work together at stalled replication forks to generate cruciform structures via fork remodeling, while Smc5/6 and Rtt107 mark the cruciforms, indicating an appropriate substrate for nuclease cleavage. As members of this pathway are conserved in humans, the uncovered mechanisms may underlie genomic instability at palindrome sites involved in disease etiology.

molecular biology↗

bioRxiv2026

Gut microbiota-derived indole-3-propionic acid preserves dorsal hippocampal catecholamines to prevent post-stroke cognitive impairment

BackgroundGut dysbiosis has been increasingly implicated in post-stroke cognitive impairment (PSCI), yet the causal contribution and therapeutic potential of gut microbiota-derived metabolites remain unclear. This study aimed to identify key microbiota-derived metabolites involved in PSCI and to elucidate their underlying mechanisms. ResultsWe found that both PSCI patients and middle cerebral artery occlusion (MCAO) mice exhibited distinct gut microbial alterations, characterized by a marked reduction in tryptophan-metabolizing bacteria and indole-3-propionic acid (IPA), a gut microbiota-derived tryptophan metabolite. Exogenous IPA administration alleviated PSCI-like phenotypes in MCAO mice. Mechanistically, IPA preserved tyrosine hydroxylase-positive (Th) fibers and catecholamine levels in the dorsal hippocampus. Further analyses showed that IPA binds to the adaptor protein Ywhab, promotes ERK activation, and enhances neuronal survival, thereby counteracting neuronal apoptosis-associated inflammation and subsequent Th fiber degeneration. ConclusionThese findings identify IPA as a gut microbiota-derived neuromodulator that mitigates PSCI by preserving dorsal hippocampal catecholaminergic transmission. IPA may therefore serve as a promising predictive biomarker and therapeutic candidate for PSCI.

neuroscience↗

bioRxiv2026

An Isoform-Centric, Structure-Aware Framework for Protein Function Prediction and Evaluation, Instantiated in 3DisoDeepPF

Understanding and accurately predicting function across protein isoforms has been a long-standing challenge with profound implications for both biological and translational research. However, most functional annotations and benchmarks remain tied to genes or reference protein sequences, leaving limited support for distinguishing protein isoforms from the same gene whose sequence, structure and domain composition may lead to different functions. To address this gap, we developed an isoform-centric framework for protein function and domain annotation that integrates a dense graph of sequence and structure similarity with protein features. We implemented this framework as 3DisoDeepPF, a graph-based multimodal model, and applied it to a breast cancer isoform atlas. Across canonical benchmarks and evaluations at isoform resolution, 3DisoDeepPF showed strong performance in predicting GO terms and Pfam domains and remained robust in tests with homology control. It further captured changes in Pfam domain composition among isoforms from the same gene, including reference-relative domain gain and loss. An evidence tracing module links predicted labels to supporting proteins in the graph, as illustrated by the calcium and integrin binding protein 1 (CIB1). Together, this study provides a framework informed by protein structure for function and domain annotation at isoform resolution, converting human protein isoform diversity into traceable functional evidence for cancer atlas interpretation.

cancer biology↗

bioRxiv2026

Learning dynamics of unsupervised deep learning reveal epoch-specific genetic architectures of brain morphology

Representation learning is an emerging paradigm for deriving phenotypes from complex measurements (e.g., imaging) for genetic discovery. However, the learning dynamics of deep neural networks, especially the evolution of representations during training, while of interest in representation learning, were insufficiently investigated in the context of genetic discovery. In this study, using a 3D convolutional autoencoder trained on T1-weighted brain MRIs UK Biobank participants, we show that its learning trajectory forms an epoch-stratified landscape of brain morphology heritability. Different training epochs capture distinct genetic architectures at comparable heritability levels. Overall, ensembling across informative checkpoints identifies more genomic risk loci than the conventional single-checkpoint approach. Interpretability analysis reveals that epoch-specific loci, including MAPT and MCPH1, map onto biologically coherent and distinct neuroanatomical signatures, identified at different stages of the training process. Our results establish learning dynamics as a novel axis for genetic discovery using unsupervised deep learning and have practical implications for any architecture that saves multiple checkpoints during training.

bioinformatics↗