Biology subjects

Zhao, Y.

Publications and source records attributed to Zhao, Y..

At least 199 records · Page 11Linked to original sources

bioRxiv2025

Predicting and Designing Red Fluorescent Protein Variants Using Sequence-to-Function Machine Learning Models

Fluorescent proteins (FPs) are widely used reporters for visualizing cellular structures and processes. Traditional wet-lab strategies for FP engineering (rational design and directed evolution) have enabled substantial improvements in photophysical performance but are limited by their requirement for deep expert knowledge or labor-intensive screening. AI-driven approaches have recently gained traction for engineering variants of green FPs, yet applications to red fluorescent proteins (RFPs) remain scarce. Here, we applied machine learning models to an RFP sequence-function dataset and trained these models to predict functional single-mutation variants of the state-of-the-art RFP mScarlet-I3. Guided by model predictions, we identified variants exhibiting red-shifted emission peaks, large Stokes shifts, or brightness comparable to the parental protein. Our findings show that even lightweight, data-efficient models can extract actionable design principles for improving RFPs. This work demonstrates the feasibility of AI-guided design for RFPs and provides a reliable benchmark for future development of more powerful AI-driven strategies for FP engineering.

molecular biology↗

bioRxiv2025

MS4MS: LLMs-driven Multi-agent System for Small-molecule Identification via LC-MS/MS

Small molecule identification is central to research fields such as drug discovery, but in complex systems like Traditional Chinese Medicine (TCM), traditional mass spectrometry analysis methods remain constrained by bottlenecks including insufficient database coverage, fragmented analysis workflows, and poor result interpretability. To address these limitations, we developed MS4MS, a large language model-driven multi-agent system that enables an end-to-end automated pipeline from raw data to small molecule identification. Validation on a public benchmark demonstrates that MS4MS achieves state-of-the-art performance in molecular formula prediction. Furthermore, its innovative small molecule identification agent enables efficient and interpretable compound elucidation. Verification using herbal extracts indicates MS4MSs outstanding performance regarding analytical coverage and the discrimination of isomers. Consequently, MS4MS offers a novel, accurate, interpretable, and high-throughput end-to-end automated strategy for small molecule identification, overcoming the analytical bottlenecks of traditional mass spectrometry in natural products and complex TCM systems.

bioinformatics↗

bioRxiv2025

Comparative evaluation of reference genomes and cell-type annotation frameworks for single-nucleus transcriptomic analysis in apple

Single-nucleus RNA sequencing (snRNA-seq) provides a powerful tool to profile cell-type-specific transcriptional programs in woody fruit trees. Yet its application in Malus domestica has been limited by heterogeneous reference genomes and the lack of validated marker genes. Here, we first established a comprehensive set of marker genes corresponding to 29 plant cell types and identified their orthologs in apple. Using these resources, we systematically evaluated six publicly available apple genome assemblies as references for snRNA-seq analysis. Our results demonstrate that reference genome selection has a major impact on mapping performance and downstream cell type annotation, with the GDDH13 assembly consistently producing the highest transcript counts per nucleus and the greatest number of identifiable transcriptional clusters (15 clusters representing 11 distinct cell types across 35,557 high-quality nuclei). To achieve robust cell type annotation, we integrated classical marker gene-based annotation with two cross-species computational frameworks -- Orthologous Marker Gene Groups (OMGs) and XSpeciesSpanner. These three approaches enabled confident identification of 11 major cell types in apple seedlings and provided cross-validation for ambiguous or previously unannotated clusters. Pseudotime analysis reconstructed the developmental trajectories of both procambial and stomatal lineages, revealing their progressive transcriptional transitions and enabling the delineation of key molecular programs underlying vascular differentiation and early guard-cell formation. Together, this work provides a reference-quality single-nucleus transcriptomic atlas for apple and underlined the critical impact of genome choice and multi-method cell-type annotation on single-cell studies in non-model crops.

bioinformatics↗

bioRxiv2025

High-confidence structural predictions of extrachromosomal DNA with ecDNAInspector

Extrachromosomal DNA (ecDNA) are circularized genomic elements that reside outside canonical chromosomes. ecDNA amplify oncogene copy number, enhance chromatin accessibility, and act as mobile enhancers through cis- and trans-regulatory interactions, collectively boosting oncogene expression. ecDNA has been implicated in tumor progression, intratumoral heterogeneity, and poor patient prognosis. Despite various lines of evidence that ecDNA promotes aggressive disease, the mechanisms and selective pressures leading to ecDNA formation and propagation remain poorly understood as are their structures. While several computational tools have been developed to infer ecDNA presence or absence from short read sequencing data, accurate identification of large or complex ecDNA structures remains challenging. Here we introduce ecDNAInspector, a novel computational framework to systematically assess the confidence of ecDNA predictions from existing inference tools. Leveraging abundant short-read whole genome sequencing (WGS) data from population-scale cohorts, we demonstrate that ecDNAInspector accurately identifies high-confidence ecDNA calls, improving interpretability and facilitating the association with clinical features. As an illustrative example, applied to a cohort of 250 breast cancers, ecDNAInspector identifies associations between ecDNA structure and molecular subgroups of disease. These findings are supported by orthogonal omic data and experimental characterization of ecDNA captured in representative cell lines. ecDNAInspector provides a scalable, data-driven approach to characterize ecDNA structure, enabling integrative studies of the clinical and biological impact of this non-mendelian mode of oncogene amplification and inheritance.

cancer biology↗

bioRxiv2025

DTCFinder: a bench-to-bits toolkit for label-free, whole-spectrum analysis of disseminated and circulating tumor cells in liquid biopsies

Disseminated (DTCs) and circulating tumor cells (CTCs) are rare but pivotal in understanding cancer metastasis and advancing liquid biopsy-based diagnostics. Traditional detection approaches, hindered by their dependency on epithelial markers or morphological features, impair the accurate identification of varied DTC/CTC phenotypes within liquid biopsies. We describe DTCFinder, a bench-to-bits, label-free method integrating high-throughput single-cell RNA sequencing and machine learning, enabling unbiased detection, in-depth molecular characterization, and precise tissue-of-origin tracing of DTCs/CTCs across various types of liquid biopsies, even with ultra-scarce DTC/CTC presence in peripheral blood and low sequencing depth. DTCFinder offers a robust and versatile toolkit for liquid biopsy-based cancer diagnostics, enhancing our understanding of cancer metastasis.

cancer biology↗

bioRxiv2025

scE2TM improves single-cell embedding interpretability and reveals cellular perturbation signatures

Single-cell RNA sequencing technologies have revolutionized our understanding of cellular heterogeneity, yet computational methods often struggle to balance performance with biological interpretability. Embedded topic models have been widely used for interpretable single-cell embedding learning. However, these models suffer from the potential problem of interpretation collapse, where topics semantically collapse towards each other, resulting in redundant topics and incomplete capture of biological variation. Furthermore, the rise of single-cell foundation models creates opportunities to harness external biological knowledge for guiding model embeddings. Here, we present scE2TM, an external knowledge-guided embedded topic model that provides a high-quality cell embedding and interpretation for scRNA-seq analysis. Through embedding clustering regularization method, each topic is constrained to be the center of a separately aggregated gene cluster, enabling it to capture unique biological information. Across 20 scRNA-seq datasets, scE2TM achieves superior clustering performance compared with seven state-of-the-art methods. A comprehensive interpretability benchmark further shows that scE2TM-learned topics exhibit higher diversity and stronger consistency with underlying biological pathways. Modeling interferon-stimulated PBMCs, scE2TM simulates topic perturbations that drive control cells toward stimulated-like transcriptional states, faithfully mirroring experimental interferon responses. In melanoma, scE2TM identifies malignant-specific topics and extrapolates them to unseen patient data, revealing gene programs associated with patient survival.

bioinformatics↗

bioRxiv2025

IgG-Bridging-Seeded Synergistic Aggregation of SARS-CoV-2 Spikes Underlies Potent Neutralization by A Low-Affinity Antibody

Mechanistic studies of viral neutralization typically prioritize high-affinity antibodies, relegating low-affinity binders to the sidelines. We report P5-1C8, a Class 1 SARS-CoV-2 antibody that exemplifies this underexplored "low-affinity yet high-potency" phenotype, retaining strong neutralization of Omicron JN.1 despite markedly weakened trimer binding (KD = 225 nM; IC50 = 0.06 nM). Structural and biophysical analyses reveal that P5-1C8 engages WT and BA.1 spikes through canonical intra-spike bivalency, but with JN.1 it induces aggregation. Using virion-like nanoparticles displaying multiple spikes, we show that IgG remains bound with no detectable dissociation and triggers pronounced aggregation. Coarse-grained molecular dynamics delineate the stepwise pathway in which weak IgG-spike contacts seed aggregation via transient inter-spike bridging. Together, these findings establish the first mechanistic framework demonstrating how weak-binding antibodies can nonetheless achieve potent neutralization through higher-order aggregation, thereby expanding the conceptual landscape of antibody function and opening new directions for antibody evaluation and design. Graphical AbstractLow-affinity antibodies are frequently disregarded in discovery pipelines. This work reports P5-1C8, a Class 1 SARS-CoV-2 antibody with weak trimer binding (KD-to-IC50 > 3,700-fold) yet potent neutralization of Omicron JN.1. Structural, biophysical, functional and coarse-grained simulations collectively demonstrate that transient inter-spike IgG bridging seeds higher-order aggregation, which in turn drives neutralization and provides a mechanistic framework. O_FIG O_LINKSMALLFIG WIDTH=199 HEIGHT=200 SRC="FIGDIR/small/691454v1_ufig1.gif" ALT="Figure 1000"> View larger version (79K): org.highwire.dtl.DTLVardef@1d4f132org.highwire.dtl.DTLVardef@1274400org.highwire.dtl.DTLVardef@e186a6org.highwire.dtl.DTLVardef@4f1870_HPS_FORMAT_FIGEXP M_FIG C_FIG

cell biology↗

bioRxiv2025

DNA-barcoded microneedle array patch enabling the profiling of in vivo spatiotemporal transcriptomics

Recent years have witnessed profound advances in spatial transcriptomics (ST). However, current ST strategies are limited to tissue sections, unable to directly profile spatiotemporal transcriptome landscapes of individual living animals. To overcome this limitation, we present acupuncture-extracted in vivo spatiotemporal transcriptomics (aivST). It is mainly based on engineering clinical-grade acupuncture gold microneedles as a spatial-barcoding array patch with high-performance DNA recognition interfaces. These interfaces first utilize nanoparticles surface deposition to increase the density of DNA probes. A freeze-thawing method is further developed to regulate interfaced DNA probes with laterally uniform distribution and vertically upright conformation, mainly by loose stretching and local concentration of these DNA probes. It increased mRNA capture with 250% and speeded up reaction (from 60 min to 15 min), compared to that without freeze-thawing surface engineering. Using this method, we explored in vivo spatiotemporal transcriptome dynamics during skin wound healing of individual mice. We found four distinct expression changes of transcripts, including continuous upregulation, continuous downregulation, initial upregulation then downregulation, and initial downregulation then upregulation. Our aivST method advances the field from sparse tissue section sampling toward in vivo measurements of biological stereotypy and variability in individual living animals.

genomics↗

bioRxiv2025

Unraveling the archaeal virosphere: diversity, functional and virus-host interactions

Archaea, the third domain of life, play critical roles in global biogeochemical cycles. However, their virosphere, particularly the proviruses which integrated into host genomes, remains largely unexplored. To systematically reveal the landscape of archaeal proviruses, we conducted large-scale mining of public and in-house genomic datasets spanning all presently known archaeal phyla. We identified 9,697 archaeal proviruses across 19 archaeal phyla and 366 families, which clustered into 9,123 viral operational taxonomic units (vOTUs). Among these, 97.2% represent novel viruses, and 81.3% could not be classified at the family level, substantially expanding the known diversity of archaeal viruses. Host range analysis revealed that many proviruses exhibit broad infectivity across archaeal lineages, with some even capable of cross-domain infection. Genomic analysis identified 178 distinct types of antiviral systems in archaeal hosts, encompassing multiple CRISPR-Cas variants and restriction-modification (RM) systems. Meanwhile, we detected 747 anti-defense genes encoded by 710 proviruses, such as anti-CRISPR and anti-RM, directly corroborating the ongoing evolutionary arms race between archaeal hosts and their viruses. Additionally, we identified 532 auxiliary metabolic genes (AMGs) within archaeal proviruses that are involved in key processes including carbon, nitrogen, and sulfur metabolism, indicating their potential to reprogram host metabolic pathways and thereby influence biogeochemical cycling. This study establishes a systematic global genomic atlas of archaeal proviruses, advancing our understanding of their distribution and diversity while laying the groundwork for future investigations into how AMG-mediated processes influence archaeal metabolism and ecosystem functions.

ecology↗

bioRxiv2025

Integrative spatial multi-omics reveals prognostic tumor niches in female genital tumors

Female genital tumors (FGTs), including ovarian, endometrial, and cervical cancers, pose a major global health challenge, yet their spatial and molecular determinants of progression and therapy resistance remain poorly understood. Here we present a large-scale integrative spatial and multi-omics atlas of FGTs, combining exome, transcriptome, methylome, and miRNA sequencing with high-resolution Visium HD spatial transcriptomics, profiling over 62 million cells from 100 tumors. We show that HPV-driven suppression of TGF{beta} signaling in cervical cancer disrupts myofibroblast (myCAF) formation and reshapes immune infiltration, while serous tumors exhibit extensive oncogene amplifications that activate MYC and PI3K pathways, linking chromosomal instability (CIN) to proliferative, immune-cold phenotypes. In high-grade serous ovarian cancer, we identify a previously unrecognized malignant domain, the Proliferating Core Desert (PCD) niche, defined by CIN-high epithelial enrichment, immune exclusion, stromal protection, and chemotherapy resistance, maintained through crosstalk with macrophages and fibroblasts. Finally, we develop POWER, a pathology foundation model that predicts PCD niche abundance from routine H&E slides, enabling robust stratification of prognosis and anti-angiogenic therapy response.

genomics↗

bioRxiv2025

High-resolution MRI Guided Whole Mouse Brain Cell Type Atlas using Deep Learning

Cell types represent groupings of cells defined by shared anatomical and functional properties. Traditional mouse brain cell atlases rely heavily on single-cell sequencing, which provides valuable molecular detail but lacks whole-brain, isotropic resolution. Diffusion MRI (dMRI) offers a complementary approach for probing cytoarchitecture and myeloarchitecture, with quantitative metrics increasingly used as biomarkers of brain development and neurodegenerative disorders. Although prior work has linked dMRI metrics with gene expression, the capacity of dMRI to directly predict cell types remains unclear. Here, we present a deep learning framework that integrates high-resolution dMRI with three-dimensional light-sheet microscopy (LSM) registered to the Allen Mouse Brain Common Coordinate Framework (CCFv3). We investigated correlations between dMRI and spatial transcriptomics-derived cell types and generated a whole-brain cell type atlas at 10 {micro}m isotropic resolution. Together, these results establish an efficient, high-resolution strategy for brain cell atlas generation and underscore the potential of advanced imaging to illuminate cellular mechanisms of the brain.

neuroscience↗

bioRxiv2025

Multi-stem-loop constrained library enables accelerated SELEX for aptamers with superior kinetics and affinity

AbstractAptamers are single-stranded nucleic acids with ligand-binding capacity as cost-effective alternatives to antibodies. However, their utility is often undermined by the inherent conformational instability of single-stranded nucleic acids, which can compromise binding kinetics and final affinity. To overcome this critical limitation, we synthesized a novel primary library incorporating multiple inlaid stem-loop structures to enhance molecular stability and rigidity. This structural constraint dramatically accelerated the selection process, allowing us to isolate aptamer candidates with superior affinity in just three rounds of SELEX (Systematic Evolution of Ligands by EXponential enrichment), a significant reduction compared to conventional methodologies. Critically, the resulting aptamer demonstrated higher affinity and faster binding kinetics compared to previously reported sequences, which directly validate our hypothesis that structural stabilization leads to high-performance aptamers. This methodology, which integrated rational conformational constraint with a high-throughput selection process, offered a generalizable strategy to efficiently select aptamers with excellent kinetic and thermodynamic performance for demanding applications such as continuous, real-time ligand detection.

biochemistry↗

bioRxiv2025

Repurposed endogenous virus-like vesicles mediate dendritic cell long-range antigen presentation and T cell activation for enhanced cancer vaccination

Dendritic cells (DCs) release extracellular vesicles (DEVs) that amplify antigen presentation while incorporating patient-derived, rapidly evolving antigens. By enriching peptide-MHC and co-stimulatory ligands orders-of-magnitude above donor-cell levels, DEVs emerge as potent vesicle-vaccines, although efficacy remains limited by unclear mechanisms. We show that DCs repurpose viral components to generate endogenous virus-like vesicles (VLVs) that preferentially carry peptide-MHC and high-density co-stimulatory ligands, intensifying and extending antigen presentation. Upon antigen exposure, Arc partners with endogenous envelope proteins to assemble VLVs that directly engage T cells and trigger intrinsic adjuvanticity via viral mimicry. Arc-/- DEVs failed to prime antigen-specific T cell responses, whereas Arc overexpression with its 5'-UTR stem-loop shifted DEVs toward VLVs that trafficked to lymphoid organs, drove rapid CD4-assisted priming and durable CD8-biased memory, suppressed melanoma, and prolonged survival. These reveal a viral-mimicry mechanism enabling long-range immune activation and support Arc+ VLVs as an antigen-agnostic vaccine for cancer immunotherapy. HighlightsO_LIArc+ VLVs intensify and extend DC antigen presentation in vivo C_LIO_LIArc+ VLVs directly engage T cells and trigger viral-mimic adjuvanticity C_LIO_LIArc-/- DEVs fail to prime antigen-specific T cell responses C_LIO_LIEngineered Arc+ VLVs drive durable CD8-biased memory and tumor control C_LIO_LIArc+ VLV vaccination provides long-range and cross-tumor protection in vivo C_LI

bioengineering↗

bioRxiv2025

The Human-In-the-Loop Drug Design Framework with Equivariant Rectified Flow

Advancements in AI-based drug design often face obstacles due to incomplete datasets, hindering progress to clinical trials. Human experts bring invaluable expertise and nuanced contextual understanding to drug design. There are two main difficulties in integrating human knowledge into the drug development process. First, human annotations are costly, and traditional machine learning algorithms require a large number of samples to be effective. Second, human experts are unable to accurately describe their expertise using natural language, nor can they precisely specify what kind of molecule is needed before seeing the generated molecules. To address these problems, we propose a new platform, called HIL-DD, for experts to infuse their experience by selecting molecules generated by AI that meet their criteria, or discarding those that do not. The core generative technology utilizes an Equivariant Rectified Flow Model (ERFM), which offers faster generation speeds than conventional diffusion models, enabling efficient human-AI collaboration. More importantly, we provide a user-friendly interface to ensure smooth and effective collaboration between human experts and AI systems. Rigorous experiments demonstrate that our system can produce 3D molecules that align with expert expectations in minimal interactive sessions. These generated molecules maintain drug-like qualities comparable to those created by current state-of-the-art models.

bioinformatics↗

bioRxiv2025

A stress-dependent TRIM28-ALKBH2 feedback loop modulates chemotherapy response

Chemoresistance to DNA-damaging agents, including platinum and alkylating compounds, limits treatment efficacy in non-small cell lung cancer (NSCLC) and is frequently associated with elevated DNA repair activity. Here, we identify a stress-responsive feedback loop between the E3 ligase TRIM28 and the demethylase ALKBH2. TRIM28 binds ALKBH2 and promotes its K48-linked polyubiquitination and proteasomal degradation, whereas ALKBH2 enhances TRIM28 transcription and protein levels, forming a feedback loop. Notably, alkylation stress induces a biphasic response: acute MMS exposure enhances TRIM28-ALKBH2 association and accelerates ALKBH2 degradation, whereas prolonged exposure promotes TRIM28 degradation, leading to ALKBH2 stabilization and transcriptional upregulation. Clinically, ALKBH2 is frequently elevated in lung adenocarcinoma and is associated with worse survival, whereas TRIM28 exhibits prognostic value specifically in chemotherapy-treated patients. Functionally, MMS-cisplatin co-treatment increased DNA damage and reduced clonogenic survival by counteracting ALKBH2-dependent alkylation tolerance, linking the TRIM28-ALKBH2 loop to chemoresistance in NSCLC.

biochemistry↗

bioRxiv2025

Developmental deviations of structural connectivity in youths with ADHD predict symptom and treatment outcomes

Attention-deficit/hyperactivity disorder (ADHD) lacks validated biomarkers to track symptom development or guide treatment selection in youth. We quantified individual deviations from normative age-related trajectories of white-matter structural connectivity (SC) across two independent cohorts: 6,687 typically developing and 1,114 ADHD scans from the ABCD study, and 292 typically developing and 482 ADHD participants from a Chinese cohort. Youth with ADHD showed pronounced SC deviations, especially in higher-order association pathways. A subset of connections showed ADHD-specific developmental reductions in deviations beyond typical developmental patterns, and these changes mediated the age-related decline in ADHD symptoms. Longitudinal within-individual analyses demonstrated that decreases in deviation over two years tracked symptom improvement. Baseline deviations also predicted 12-week treatment response to atomoxetine, but not methylphenidate, and follow-up imaging revealed treatment-related reductions in deviation. Together, these findings identify SC deviation as a robust developmental biomarker with prognostic and theragnostic relevance for ADHD, supporting precision care through risk stratification, individualized pharmacotherapy selection, and objective monitoring of treatment effects.

neuroscience↗

bioRxiv2025

CDH3 as a Novel Therapeutic Target in Basal-like Double-Negative Prostate Cancer

PurposeBasal-like (also known as double-negative) prostate cancers are aggressive tumors that lack effective targeted therapies. We aimed to delineate the role of CDH3 (P-cadherin) in basal-like prostate cancer and evaluate CDH3-directed therapeutic strategies. MethodsWe integrated genetically engineered mouse models (GEMMs) of prostate cancer, bulk and single-cell transcriptomic analyses, and a suite of in vitro and in vivo experiments. CDH3 expression and associated signaling pathways were examined in Pten/Apc double-knockout mouse prostates and human datasets. Functional studies included antibody-drug conjugate (ADC) cytotoxicity assays and the development of chimeric antigen receptor (CAR) T cells targeting CDH3, tested in prostate cancer cell lines and xenograft models. ResultsPten/Apc double deletion in prostate GEMMs led to highly aggressive tumors with markedly elevated CDH3 expression and enrichment of non-canonical WNT signaling components. Transcriptomic analyses of patient-derived prostate tumors confirmed that CDH3 is significantly upregulated in basal-like prostate cancer subtypes relative to luminal subtypes. Single-cell RNA sequencing revealed CDH3 expression predominantly in basal epithelial cells. Mechanistically, we found that active YAP1 signaling and a WNT5A-ROR2 non-canonical WNT axis drive CDH3 expression. Targeting CDH3 with a CDH3-specific ADC induced potent, antigen-dependent killing of CDH3 prostate cancer cells in vitro and significantly suppressed tumor growth in in vivo metastatic prostate cancer models. Likewise, CDH3-targeted CAR T cells specifically recognized and lysed CDH3-expressing prostate tumor cells while sparing CDH3-negative cells, leading to tumor regression and improved survival in mouse models, especially when combined with PD-1 checkpoint blockade. ConclusionsCDH3 is a key marker and functional driver of basal-like prostate cancer. Therapeutic strategies leveraging CDH3, including ADCs and CAR T cells, demonstrate strong preclinical efficacy, supporting the development of CDH3-targeted treatments to overcome resistance in aggressive prostate cancer.

cancer biology↗

bioRxiv2025

Extracting task-relevant preserved dynamics from contrastive aligned neural recordings

Recent work indicates that low-dimensional dynamics of neural and behavioral data are often preserved across days and subjects. However, extracting these preserved dynamics remains challenging: high-dimensional neural population activity and the recorded neuron populations vary across recording sessions. While existing modeling tools can improve alignment between neural and behavioral data, they often operate on a per-subject basis or discretize behavior into categories, disrupting its natural continuity and failing to capture the underlying dynamics. We introduce Contrastive Aligned Neural DYnamics (CANDY), an end-to-end framework that aligns neural and behavioral data using rank-based contrastive learning, adapted for continuous behavioral variables, to project neural activity from different sessions onto a shared low-dimensional embedding space. CANDY fits a shared linear dynamical system to the aligned embeddings, enabling an interpretable model of the conserved temporal structure in the latent space. We validate CANDY on synthetic and real-world datasets spanning multiple species, behaviors, and recording modalities. Our results show that CANDY is able to learn aligned latent embeddings and preserved dynamics across neural recording sessions and subjects, and it achieves improved cross-session behavior decoding performance. We further show that the latent linear dynamical system generalizes to new sessions and subjects, achieving comparable or even superior behavior decoding performance to models trained from scratch. These advances enable robust cross-session behavioral decoding and offer a path towards identifying shared neural dynamics that underlie behavior across individuals and recording conditions. The code and two-photon imaging data of striatal neural activity that we acquired here are available at https://github.com/schnitzer-lab/CANDY-public.git.

neuroscience↗