Gene expression and cancer classification

This paper compares topic modeling methods on transcriptomic data to identify tissue structure and distinguish healthy from cancerous tissues, showing their effectiveness as embedding spaces for classification.

This study identifies key factors affecting field transcriptome model accuracy through simulation and validation with real data, showing gene expression patterns, sample size, and environmental coverage are most influential.

This paper introduces FS-PLS, a feature selection method that generates minimal disease signatures from omics data without requiring library size normalization, achieving smaller and comparable performance to existing methods.

This study evaluates unsupervised variable selection methods for high-dimensional omics data and introduces TRIM-IT, a tool that identifies glioblastoma subtypes with distinct molecular profiles and survival outcomes.

This study identifies a small set of transcription factors and microRNAs that can predict genome-wide gene expression across multiple cancer types with high accuracy.

This study benchmarks 35 AI classifiers across 153 cancer datasets to evaluate their performance in molecular subtype classification, identifying optimal data configurations and model types for different omics modalities and cancer types.

This study evaluates how cold ischemia time affects gene expression in normal colon mucosa, identifying transcriptomic changes in hypoxia, EMT, and apoptosis pathways that may mimic malignancy.

The paper introduces TCGEx, a web-based tool for analyzing cancer transcriptomics data, identifies microRNAs and cytokines linked to immunotherapy response in melanoma.

This study uses machine learning to identify novel shared targets in glioblastoma and osteosarcoma, revealing C2ORF72 as a potential therapeutic and diagnostic target.

RCE-IFE reduces feature set size and improves model performance by recursively eliminating low-scoring feature clusters and intra-cluster features in a supervised setting.

This paper develops OncoMark, a neural multi-task learning framework to predict cancer hallmark activities from transcriptomics data, achieving high accuracy across internal and external datasets.

The paper compares binary and rank-based methods for identifying differentially expressed genes using Bayesian statistics on a yeast dataset, highlighting issues with traditional thresholds and advocating for rank-based approaches.

This paper introduces DGBCox, a stable prediction method for breast cancer prognosis under data distribution shift, showing improved accuracy and stability compared to existing methods.

The paper identifies invariant feature genes (IFGs) using a symmetric machine learning approach to distinguish healthy from disease states and finds drug targets in cancer IFGs.

This paper introduces a Bayesian framework for differential gene expression analysis that ranks genes based on statistical evidence rather than binary classification, improving transparency and computational efficiency.

MOBER integrates transcriptomic data from cancer cell lines, xenografts, and clinical tumors using deep learning to remove batch effects and improve clinical translatability.

This study identifies biomarker genes and pathways contributing to lung cancer health disparities using a classification framework with SHAP to interpret disparities between racial and sex groups.

edgeR v4 adds a continuous negative binomial for fractional counts, bias-corrected quasi-likelihood for small counts, C-based fitting, and new transcript, exon, methylation and fold-change-threshold analyses.

This paper introduces JECCA, a method for identifying shared biological signals across gene expression datasets, enabling disease subtyping, biomarker discovery, and cross-species analysis.

The paper develops and evaluates methods for optimizing k in k-nearest-neighbor feature selection for imbalanced data, showing improved performance on RNA-Seq data from Major Depressive Disorder.

This paper introduces phenotype-driven data augmentation methods for transcriptomic data, showing improvements in patient stratification over existing approaches.

The paper introduces ComBat-ref, a batch effect correction method for RNA-seq data that improves sensitivity and specificity over existing methods.

The paper predicts plant phenotypic traits using RNA-seq datasets and finds that few hundred samples suffice for tissue classification but thousands are needed for age prediction.

The paper develops a pre-trained transformer model to predict cancer tissue of origin from bulk gene expression data, achieving high accuracy across 32 tissue types.

The paper introduces MICA, a method to detect multi-class biomarkers across multiple transcriptomic studies, showing improved accuracy and biological insights in both bulk and single-cell data.

The paper introduces CPSM, an R package that provides an automated machine learning pipeline for predicting individual cancer patient survival probabilities using multi-omics data.

The paper introduces a directional feature importance metric for decision tree classifiers and evaluates its performance on cancer proteomics data using multiple algorithms.

MAGE identifies aberrantly expressed genes by analyzing entire expression profiles rather than individual genes, outperforming conventional differential expression and density-based clustering methods.

The paper analyzes gene expression dynamics in normal and tumor tissues, identifying N-genes and T-genes that mark transitions to malignancy in prostate, lung, and liver cancers.

This paper adapts a pre-trained transformer model for bulk tumor data to predict patient survival outcomes, achieving strong correlations and outperforming traditional methods.