Gene expression and cancer classification

This paper introduces a statistical framework for system-level synthesis that learns and predicts biological processes from high-dimensional data while identifying participating features.

This paper introduces variable-sparsity matrix and tensor factorization algorithms for cancer genomics to improve biological interpretability and modeling of cancer subtypes and pathways.

The paper benchmarks 2,234 transcriptomic harmonization approaches on B-cell lymphoma data, finding ComBat excels in cross-batch prediction quality.

This study shows that surrogate variable analysis (SVA) can lead to overfitting and reduced cross-cohort reproducibility in neuropsychiatric gene expression analyses.

broadSeq benchmarks differential gene expression methods and shows DELocal identifies distinct genes by considering genomic neighbourhood, particularly for highly expressed genes.