Gene expression and cancer classification

This study develops a machine learning framework to predict CRC stage transitions using tissue-derived transcriptomic signatures and evaluates their performance in blood-based datasets.

GR-SAFS identifies a 20-gene prognostic signature from TCGA-LUAD data with high concordance index, outperforming other methods in external validation.

vDeepInsight introduces a 3D voxel carrier for tabular data that improves feature neighborhood representation, leading to better performance in gene expression tasks.

This study introduces a complex-valued representation for time-series gene expression data, showing it captures distinct gene-gene similarities and temporal programs comparable to conventional methods.

TifBERT is a self-supervised foundation model for bulk RNA-seq that learns normalization-robust representations without expression discretization or external embeddings, achieving high accuracy and capturing pathway-level biology.

The paper identifies how batch correction methods can distort gene-gene correlations and introduces a metric to quantify this issue.

This paper reviews multivariate methods for integrating histological images and gene expression data, showing they capture distinct yet complementary aspects of the underlying information.

ROTS 2.0 improves statistical modeling in omics studies by optimizing reproducibility across various experimental designs and provides open-source implementations in R and Python.

The paper benchmarks machine learning classifiers on TEP RNA-seq data to classify cancer types, finding Multinomial Logistic Regression achieves the highest performance.

The paper introduces a scale mixture model for metatranscriptomics data to improve differential expression analysis in inflammatory bowel disease, identifying new candidate genes not found by Gaussian methods.

PXN improves cross-platform gene expression data integration by removing technology-specific biases and enhancing differential expression analysis.

The paper introduces STPCaT, an isoform-centric analysis of solid tumors, revealing cancer-specific and normal-specific isoform groups and identifying new cancer-testis antigens with diagnostic potential.

TumorArchetypeR develops a modular framework to derive signature-based tumor subtypes, resolving fibrotic tumor heterogeneity and identifying clinically relevant subtypes with robust classifiers.

The paper benchmarks normalization and differential expression methods for miRNA-seq using a realistic dataset, finding RPM normalization and edgeR perform best.

TmSNet predicts tumor cell-specific total mRNA expression from multi-omic data, achieving high accuracy across cancer types and external cohorts.

This paper integrates multiple transcriptomics assays to identify cancer biomarkers, finding GBP4 as a predictor of recurrence and immune remodeling.

The paper identifies a universal molecular point of no return in cancer progression by analyzing gene coordination patterns across 8,314 tumors, revealing shared survival predictors across multiple cancer types.

This paper introduces a Bayesian method to align transcript dynamics across different time scales and species, enabling comparative analysis of gene expression patterns.

GeneLift recovers prognostic gene signatures from underpowered transcriptomic datasets by augmenting data with Gaussian mixture modeling and validating results with EvidenceClock.

The paper identifies stable housekeeping genes to improve machine learning model performance by reducing data leakage in transcriptomic data from kidney transplant recipients.

RNABag is a transcriptome foundation model that generalizes across biopsy modalities, achieving superior performance in cancer detection and patient stratification.

X-Plat transforms expression and methylation data between microarray and sequencing platforms using polynomial regression, achieving lower RMSE than existing methods across multiple species.

The paper introduces pipeML, a framework for leakage-free machine learning in omics data by enabling fold-aware cross-validation, improving model evaluation accuracy.

EES-Transformer V2 classifies tissues and learns gene representations from extreme expression sets, achieving high accuracy and revealing tissue-specific gene interactions.

The paper introduces BatchQC, an R package for assessing and correcting batch effects in high-throughput data with reproducible tools and workflows.

DEPower provides a web tool for power analysis in RNA-seq experiments based on DESeq2 assumptions, enabling researchers to determine sample size requirements for detecting differential expression.

Fuzzifier* provides a differential analysis framework for multi-omics data, identifying reliable and sensitive features by analyzing value distributions and foldchange patterns across groups.

The paper introduces RUVprps, a software package for normalizing RNA-seq data by removing unwanted variation, offering improved performance and features for analysis.

This paper introduces DFE, a feature engineering method that improves biological classification by reducing overfitting and enhancing interpretability, showing superior performance on RNA-sequencing datasets.

This study shows that scale misspecification in differential expression analyses leads to poor false discovery rate control and provides guidance on choosing appropriate scale uncertainty.