Pengyi Yang

The paper identifies a transitional stage in breast cancer metastasis characterized by malignant epithelial diversification and inflammatory neutrophil remodeling, linking these to poor clinical outcomes.

This paper introduces Phenoverse, a deep learning framework that learns disease state representations from single-cell transcriptomic data, enabling disease prediction and severity assessment using binary labels.

Hydra is a deep generative framework for single-cell omics data that improves cell type annotation and feature selection, particularly for rare populations.

This study identifies PRDM9 as a factor promoting survival of drug-tolerant persister cells in glioblastoma through metabolic rewiring and cholesterol biosynthesis.

This study evaluates how large language models generate summaries of bioinformatics analyses, finding they produce readable and safe summaries but often misinterpret visualizations and lack novel insights.

SwitchClass identifies attenuated and escalated molecular features across biological states using a label-switch classification framework, revealing directional changes in proteomics, phosphoproteomics, and single-cell transcriptomes.

CLUEY improves cell type detection and clustering in single-cell omics by integrating biological knowledge to guide clustering and enhance interpretability.

This paper benchmarks single-cell methods by analyzing 282 papers, highlighting challenges in combining benchmarking studies and preventing fatigue.

The paper develops a framework to characterize and design gene panels for spatially resolved omics, showing that both computational and manually curated panels have strengths in capturing cell types and reducing redundancy.

Refate identifies chemical compounds targeting trans-regulatory networks to enable cellular conversion by integrating single-cell data and drug databases.

The paper identifies mesendoderm progenitor cells in mouse embryos and shows their contribution to endoderm and mesoderm lineages through lineage tracing and analysis of Mixl1 function.

The paper introduces BrainSTEM, a multi-resolution mapping framework to evaluate midbrain differentiation protocols by comparing them to a fetal brain atlas, revealing off-target cell types in most published models.

This study benchmarks single-cell multimodal omics integration methods across multiple tasks to guide method selection based on data characteristics and study goals.

ThunderBolt provides an interactive web platform for analyzing large proteomics datasets through guided exploration, preprocessing, and statistical analysis.

This study evaluates blastoid models of early human development by comparing them to a reference map of human blastocysts, identifying their lineage coverage and differentiation progression.

SpaNorm improves spatial transcriptomics data normalisation by separating spatial library size effects from biological signals, enhancing domain and spatially variable gene identification.

This study evaluates spatially variable gene detection methods across multiple spatial transcriptomics datasets to assess their performance, reliability, and practicality.

The paper develops SnapKin, a deep learning ensemble model for kinase-substrate prediction from phosphoproteomics data, achieving the best performance in this task.

SnapCCESS improves clustering of multimodal single-cell omics data by integrating multiple data modalities using an ensemble deep learning framework.

This study benchmarks analytical combinations for predicting COVID-19 outcomes using scRNA-seq data, finding that ensemble learning and multi-view features improve performance.

The paper introduces a statistical method to analyze highly multiplexed imaging data and identifies distinct tissue architectures across three high-parameter assays.

Matilda is a multi-task learning method that integrates single-cell multimodal omics data by performing data simulation, dimension reduction, cell type classification, and feature selection.

This study uses phosphoproteomics to show that insulin resistance involves rewired signaling networks and widespread dysregulation of GSK3 activity.

This study evaluates similarity metrics for clustering in situ imaging cytometry data and proposes FuseSOM, an ensemble algorithm that improves clustering performance.

The paper creates an iPSC differentiation atlas and identifies TMEM88 as a regulator of mesendodermal development and blood pressure.

scFeatures creates interpretable molecular representations of single-cell and spatial data to predict disease outcomes using diverse feature types.

This study benchmarks retinal organoid protocols by comparing their cellular and molecular fidelity to fetal and adult human retinal tissues.

scREMOTE predicts regulatory gene relationships and builds a computational model for cell reprogramming using multimodal single-cell data, showing improved long-term prediction of transcription factor effects.

Cepo identifies cell identity by analyzing differentially stable genes from scRNA-seq data, improving cell-type characterization and spatial mapping.

The paper shows that bivalent chromatin protects reversibly repressed genes from irreversible silencing by maintaining H3K4me3 at promoters, which prevents de novo DNA methylation in adult cancers.