Gene expression and cancer classification

The paper introduces 2s-LCA, a method for joint analysis of multiple biological data sets, and applies it to human brain development data to identify shared gene expression patterns.

PairGP models longitudinal gene expression data from paired multi-condition studies to identify condition groups with different dynamics.

The paper identifies cancer-related genes by analyzing changes in gene expression variability and distribution, finding distinct sets of genes missed by traditional differential expression analysis.

This study compares neural network architectures for multi-cancer classification using miRNA, protein, and mRNA data, finding that shallow networks outperform deeper models.

oCEM detects and analyzes overlapping co-expressed gene modules using decomposition methods and a permutation-based approach to select principal components, outperforming existing techniques.

PanClassif improves cancer classification from single-cell RNA-seq using machine learning by selecting effective genes and applying smoothing and oversampling techniques, achieving better performance than existing methods.

reComBat removes batch effects from large-scale, multi-source omics data while preserving biological variation, outperforming existing methods for public gene expression data of Pseudomonas aeruginosa.

The paper introduces XAE4Exp, an explainable autoencoder for gene expression data, identifying non-differentially expressed genes and cancer-related pathways using TCGA data.

MFmap predicts cancer subtypes and similarity between cell lines and tumours using a semi-supervised generative model with high accuracy.

This paper introduces a restricted maximum-likelihood method to estimate latent variance components in gene expression data, improving efficiency and accuracy over gradient-based approaches.

This paper introduces OptICA, a method for selecting optimal dimensionality in ICA of transcriptomic data, showing it outperforms existing methods across multiple datasets.

This paper introduces CLA, a method to identify cell-type-specific genes and alternative promoters by integrating Cell Ontology and random forest classification with stratified bootstrapping.

The paper uses principal component analysis to identify differentially expressed genes in cells exposed to a collagen-glycosaminoglycan mesh compared to controls, finding enriched biological terms and better performance than linear regression.

The paper introduces DFC, a method using adaptive LASSO logistic regression to identify a small subset of genes that distinguish cell populations, addressing issues with P-value inflation in single-cell transcriptomics.

The paper applies sparse k-means clustering to human brain development data to identify age-related clusters and reveal lifespan-stage changes in the visual cortex.

GEDI integrates transcriptomic data from microarray and NGS platforms by removing batch effects and re-annotating data, verified through PCA and logistic regression.

The paper introduces a sparse orthogonal projection method for dimensionality reduction, applied to miRNA expression analysis and cancer prediction, showing competitive classification performance.

The paper evaluates the impact of two-step batch correction on gene expression analysis and proposes strategies to mitigate its effects on statistical inference.

The paper identifies four critical genes with high accuracy in lung cancer subtypes and functional effects, showing robustness across different cohorts and revealing new biological mechanisms.

GEOexplorer is an R/Bioconductor package that enables interactive and reproducible microarray gene expression analysis with visualisation tools.

This paper introduces a causality-inspired feature selection method, CLAND, to improve accuracy and stability in cancer biomarker discovery from imbalanced omics data.

This study uses a variational autoencoder to encode tumor transcriptomes and combines it with XGBoost to predict chemotherapy response across five cancer types, achieving better performance than PCA or raw data in four types.

SGI identifies clinical subgroups in omics datasets using hierarchical clustering and association testing, demonstrated on a metabolomics dataset from a type 2 diabetes study.

Hobotnica estimates molecular signature quality by assessing their ability to stratify data and assign significance scores for comparing different groups.

This paper demonstrates that using stratified loss functions in deep learning improves survival prediction in multi-dataset settings by accounting for varying baseline hazards.

The paper applies matrix factorization methods to reduce high-dimensional multi-omic data in systems pharmacology, identifying predictive gene signatures for drug response.

This paper introduces GIM-CNNs, a framework using gene interaction matrices and convolutional neural networks to detect biological entities, outperforming standard methods in identifying disease-relevant factors beyond individual genes.

This study compares five methods for analyzing gene expression in multi-sampled prostate tumors and benchmarks linear mixed models against generalized linear models for differential expression analysis.

This study combines ensemble systems biology feature selection with a bimodal deep neural network to improve breast cancer prognosis prediction, achieving high AUC and validated survival differentiation.

The paper uses deep neural networks to predict gene perturbation from transcriptional profiles, achieving 74.9% accuracy but with limited generalizability due to data consistency issues.