Zhi Wei

FedEdgeR enables federated differential gene expression analysis while preserving privacy, outperforming meta-analysis methods and matching pooled edgeR results.

The paper shows that naive fusion of sequence and graph modalities in TCR-peptide binding prediction can degrade performance and introduces TRACE, a contrastive alignment method that stabilizes multimodal learning.

CellPilot improves single-cell annotation accuracy by integrating small language models into a structured workflow, achieving performance close to larger models while maintaining transparency.

CoLa-VAE improves denoising and representation learning in single-cell data by integrating cell-cell communication information, enhancing downstream analyses like differential communication detection and spatial deconvolution.

This study uses single-cell lineage tracing to map clonal and transcriptional dynamics in melanoma metastasis, identifying distinct transcriptional programs and spatial organization patterns.

This study identifies drug-tolerant persister states in melanoma using MeRLin, showing their role in resistance and tumor recurrence.

This study identifies ZL0580 as a BRD4-targeting small molecule that suppresses HIV in vivo through epigenetic mechanisms, showing potent antiviral effects in a humanized mouse model.

This study evaluates a pre-trained model's ability to predict eye gaze from fMRI data during movie watching, finding limited individual accuracy but reliable group-level results for brain activity related to eye movements.

This paper introduces TIDE, a dual-encoder framework using cross-attention fusion to predict TCR-pMHC binding, achieving better accuracy than existing methods on benchmark datasets.

This study examines how opioid abuse affects cell-cell communication in the ventral tegmental area using snRNA-seq data, finding reduced overall communication but enhanced NRXN signaling in specific pathways.

This study identifies dysregulated gene regulatory networks in opioid use disorder and HIV using snRNA-seq data from brain samples, revealing altered FOS-centered networks and potential therapeutic targets.

scDILT integrates single-cell datasets while preserving reference clustering patterns and mapping new cells to annotated clusters, outperforming existing methods on simulated and real data.

This study uses large language models to analyze EBV-related publications, identifying key antigens and pathways involved in EBV-mediated diseases and validating their roles in specific cancers.

This study identifies molecular mechanisms of resistance to immune checkpoint blockade in metastatic melanoma using proteogenomic and spatial analyses of tumor and immune microenvironment data.

The paper introduces strategies to detect gene-gene interactions in structured populations by addressing confounding from population structure using model-based multifactor dimensionality reduction.

Genetic correlations across 25 brain disorders (215,683 cases) reveal broad psychiatric common-variant sharing, largely distinct neurological disorders, and migraine as the only significant cross-category link.