Zhicheng Ji

Halo improves whole-cell segmentation in spatial transcriptomics by integrating nuclear morphology and RNA transcript distribution, outperforming existing methods across diverse tissues.

L-STAR improves spatial domain detection by combining visual LLM guidance with consensus-based integration of multiple methods across diverse datasets.

This paper develops a vascularized tumor organoid platform to model patient-specific immunotherapy responses and identifies hyperangiogenic signaling as a barrier to T cell infiltration.

Lense optimizes single-cell omics preprocessing by using language models to select the best pipeline based on visual comparisons of low-dimensional representations.

MUSTARD reduces dimensionality of multi-sample scRNA-seq data using trajectory guidance, revealing sample-level heterogeneity and associated gene markers.

PreTSA efficiently models temporal and spatial gene expression patterns in large single-cell and spatial transcriptomics datasets, matching state-of-the-art results with reduced computational time.

This study evaluates large language models' ability to solve biomedical data science challenges using a classroom experiment on Kaggle, finding they produce competitive results despite not topping leaderboards.

Pseudotimecascade visualizes multi-gene expression cascades along pseudotime and links them to biological functions in hematopoietic stem cell differentiation.

The paper introduces TWiST, a method for transcriptome-wide association studies at cell state resolution using single-cell eQTL data, showing improved power in detecting dynamic gene-disease associations.

DeepScence identifies senescent cells in single-cell and spatial transcriptomics data, outperforming existing methods.

This paper benchmarks large language models for genomic knowledge using GeneTuring, finding that a custom GPT-4o setup with NCBI APIs performs best.

Vispro improves imaging analysis for Visium spatial transcriptomics by automating image processing tasks and enhancing downstream analysis performance.

This study evaluates large multimodal models for one-shot learning and interpretability in biomedical image classification, showing their superiority over traditional methods.

This study evaluates 18 cell segmentation methods on light microscopy and fluorescence images, finding attention-based models perform best and provides a pre-trained resource for segmentation.

SifiNet identifies gene sets and annotates cells without clustering, showing better performance than existing methods on multiomic data.

This study evaluates GPT-4 for cell type annotation in single-cell RNA-seq, showing it can accurately annotate cell types using marker gene data and reducing manual effort.

The paper examines the relationship between T cell receptor sequences and gene expression levels, finding correlations and predictive associations.

Lamian is a statistical framework for analyzing pseudotime differences across multiple scRNA-seq samples, improving generalizability by accounting for sample variability.

GeneSegNet improves cell segmentation by integrating gene expression and imaging data, outperforming methods that neglect or underuse these inputs.

Smmit integrates multi-sample single-cell multi-omics data across samples and modalities, effectively removing batch effects and improving computational efficiency.

The paper analyzes DNA methylation entropy in human and mouse datasets, finding associations with sequence features, developmental patterns, and gene expression variability.

The paper identifies cis-regulatory elements in dorsal root ganglion neurons that show differential chromatin accessibility and TF binding after nerve injury, linking them to chronic pain.

Paella decomposes spatial heterogeneity in cell trajectories from spatial transcriptomics data, identifying spatially varying genes linked to tumor progression.

Palo optimizes color palettes for single-cell and spatial genomic data to improve visualization by assigning distinct colors to spatially neighboring clusters.

findPC automatically selects the optimal number of principal components for single-cell data using six methods, with the perpendicular line method performing best on real RNA-seq data.

The paper shows GLI3 repression does not prevent precocious activation of HH target genes or regulate anterior-posterior limb polarity.

TreeCorTreat analyzes and visualizes phenotype-cell type associations in multi-sample single-cell RNA-seq data using a tree-based correlation approach.

The paper identifies transcriptomic and epigenetic changes in rat dorsal root ganglia under two chronic pain models, revealing shared and distinct chromatin accessibility alterations.

This study evaluates 18 scRNA-seq imputation methods using cell line and tissue data, finding significant variability in performance and recommending caution in downstream analyses.

SCATE improves the accuracy of estimating cis-regulatory element activities in single-cell ATAC-seq data by integrating co-activated elements, similar cells, and regulome data.