Peng Wei

Mendel is a foundation model trained on human whole genome sequences to prioritize regulatory variants and improve gene expression prediction.

The paper introduces an M-learner approach for detecting heterogeneous mediation effects in high-dimensional omics data, showing it identifies subgroup-specific biological pathways in population studies.

This study identifies stemness features in luminal hormone-responsive cells within DCIS that correlate with invasive progression and shows that targeting FOXA1 reduces tumor growth.

TUSCAN improves tumor region identification in spatial transcriptomics by integrating copy number variation profiles with histopathological images.

This paper compares statistical methods for differential gene expression in spatial transcriptomics, finding that the GEE framework better controls false positives than the Wilcoxon test.

MELODY quantifies total mediation effects in logistic regression for high-dimensional mediators and binary outcomes, outperforming traditional methods in non-rare disease scenarios.

This study benchmarks five DNA foundation models across genomic tasks, finding that mean token embeddings improve classification performance and specialized models outperform general ones in gene expression and QTL prediction.

The paper introduces an R2 total mediation effect measure for binary traits in case-control studies with high-dimensional mediators, showing 89% of BMI's effect on CHD is mediated by metabolomics.

The paper develops a meta-analysis framework for estimating total mediation effects with high-dimensional omics mediators using summary statistics, applied to TOPMed data to study gene expression's role in blood pressure and HDL cholesterol variation.

MASH identifies high-dimensional mediators in survival analysis, showing cigarette smoking mediates 51.1% of coronary heart disease risk via five metabolites.

Spatial IMIX integrates spatially resolved multi-omics data with sample subtypes, modeling spatial and inter-data-type correlations to identify cancer-initiating gene activities.

The paper introduces a cross-fitted estimation procedure for R2-based mediation effects in high-dimensional mediators, showing improved computational efficiency while maintaining statistical accuracy.

This study identifies molecular changes in bladder mucosa that initiate cancer, revealing inflammation's role and mutation patterns linked to early carcinogenesis.

IMIX integrates multiple omics data types using a multivariate mixture model to improve FDR control and reduce misclassification rates compared to single-omics analyses.

The paper extends a mediation effect measure to high-dimensional omics data and applies it to the Framingham Heart Study, finding that gene expression explains over half of aging-related systolic blood pressure variation.

This study identifies distinct genomic and molecular features of sarcomatoid bladder cancer, linking EMT dysregulation to aggressive behavior and immune infiltration.

FunSPU is a multiple functional annotation-based association test for WGS data that adapts to annotations and variants, identifying LDL-associated loci missed by existing methods.