Biology subjects

Xun, D.

Publications and source records attributed to Xun, D..

5 recordsLinked to original sources

bioRxiv2026

V3Cell: A Vision-Guided Virtual 3D Cell Framework for Phenotypic Modeling and Perturbation Prediction

Predicting how organoids respond to chemical perturbations is central to disease modeling and drug discovery. Existing virtual cell models operate at the single-cell level, producing static endpoint predictions from destructive assays. This leaves a critical gap at the organoid scale, where biological identity is defined by tissue-level architecture and continuous developmental dynamics rather than single-cell features. Here we introduce V3Cell, a vision-guided framework that constructs in silico surrogates of organoids directly from non-invasive bright-field microscopy. A foreground-aware model constructs static virtual 3D cells across colon, stomach, and lung organoid lineages. These virtual 3D cells closely match real samples across distributional metrics, micro-texture, and lineage-specific morphometrics, with small effect sizes for most descriptors. A temporal module further predicts developmental fate from as few as six early-frame observations and models fate-conditioned spatiotemporal trajectories that closely recapitulate real perturbation responses. V3Cell requires no omics profiling or fluorescent labeling, establishing a non-invasive brightfield-based paradigm for organoid-scale perturbation prediction. Our code and data are publicly available at https://github.com/Laineyoulu/V3Cell.

bioinformatics↗

bioRxiv2026

AI-driven discovery and engineering of human endogenous nanocage proteins for mRNA delivery

Safe and effective gene delivery remains a central challenge for therapeutic applications. While non-viral and viral vectors have enabled substantial progress, their reliance on non-human components often triggers immune responses, limiting their use in chronic treatments. Here, we developed DeepDelivery, an artificial intelligence-driven platform to repurpose human proteins for mRNA delivery. An unbiased screening of the human proteome nominated 512 candidates, with experimental validation confirming that 80% of top-ranked hits form mRNA-encapsulating particles and mediate efficient functional delivery in human cells without provoking detectable inflammation. Notably, multiple tripartite motif (TRIM) family proteins, typically linked to antiviral responses, exhibited strong assembly and delivery activity. Quantitative analysis and interpretation of the model revealed structural domains that govern nanocage formation, enabling domain-guided engineering of TRIM25 variants with enhanced function. Our work establishes a generalizable framework for discovering human-derived delivery vehicles and provides a path toward programmable, non-immunogenic mRNA therapeutics.

synthetic biology↗

bioRxiv2025

Systematic evaluation of blood contamination in nanoparticle-based plasma proteomics

Circulating blood proteomics enables minimally invasive biomarker discovery. Nanoparticle (NP)-based circulating plasma proteomics studies have reported varying number of proteins, ranging from ca 2000 to 7000 but its not clear whether higher protein number is more informative. Here, we first develop OmniProt - a silica-NP workflow for plasma proteomics which is optimized through systematic evaluation of NP types and protein corona formation parameters. Next, we present an Astral spectral library for 10,109 protein groups expressed in human plasma. Using Astral with a throughput of 60 sample-per-day, OmniProt identifies ca 3000 to 6000 protein groups from human plasma. Notably, we found that platelet/erythrocyte/coagulation-related contamination artificially elevates protein identifications and compromises quantification accuracy for circulating plasma proteins in NP-enriched samples. Through controlled contamination experiments, we identified and validated protein biomarkers for contaminations of platelet, erythrocyte, and coagulation, respectively, in NP-based plasma proteomics experiments. We further developed an open-access software Baize for accessing plasma proteomics contamination. Finally, we validated our OmniProt workflow and Baize software in a clinical cohort of 193 patients with CT-indistinct benign nodules or early lung cancers. On average, we identified ca 4400 protein groups for each plasma sample without significant contamination. Five samples were flagged out as contaminated. In conclusion, this study reveals that contamination of plasma samples significantly alters the protein identification and quantification in NP-based plasma proteomics experiments, and presents an open-access software Baize to evaluate plasma proteome contamination.

biochemistry↗

bioRxiv2023

Microsnoop: a generalist tool for the unbiased representation of heterogeneous microscopy images

Microscopy image profiling is becoming increasingly important in biological research. Microsnoop is a new deep learning-based representation tool that has been trained on large-scale microscopy images using masked self-supervised learning, eliminating the need for manual annotation. Microsnoop can unbiasedly profile a wide range of complex and heterogeneous images, including single-cell, fully imaged, and batch-experiment data. Its performance was evaluated on seven high-quality datasets, containing over 358,000 images and 1,270,000 single cells with varying resolutions and channels from cellular organelles to tissues. The results show that Microsnoop outperforms previous generalist and even custom algorithms, demonstrating its robustness and state-of-the-art performance in all biological applications. Furthermore, Microsnoop can contribute to multi-modal studies and is highly inclusive of GPU and CPU capabilities. It can be easily and freely deployed on local or cloud computing platforms.

bioinformatics↗

bioRxiv2021

Scellseg: a style-aware cell instance segmentation tool with pre-training and contrastive fine-tuning

Deep learning-based cell segmentation is increasingly utilized in cell biology and molecular pathology, due to massive accumulation of diverse large-scale datasets and excellent progress in cell representation. However, the development of specialized algorithms has long been hampered by a paucity of annotated training data, whereas the performance of generalist algorithm was limited without experiment-specific calibration. Here, we present a deep learning-based tool called Scellseg consisted of novel pre-trained network architecture and contrastive fine-tuning strategy. In comparison to four commonly used algorithms, Scellseg outperformed others in average precision and Aggregated Jaccard Index on three disparate datasets. Interestingly, we found that eight images are sufficient for model tuning to achieve satisfied performance based on a shot data scale experiment. We also developed a graphical user interface integrated with functions of annotation, fine-tuning and inference, that allows biologists to easily specialize their self-adaptive segmentation model for analyzing images at the single-cell level.

bioinformatics↗