Chloé‐Agathe Azencott

This study improves the estimation of evolutionary distances and tree inference by accounting for substitution rate heterogeneity across prokaryotic genomes.

This paper applies the knockoff procedure to transcriptomic data for variable selection, finding it outperforms existing methods but is conservative in real datasets.

The paper applies network-based methods to genome-wide biobank data to discover genetic associations in psoriasis, identifying known genes and novel MRPL family genes.

This paper introduces SMuGLasso, a method to address population stratification in GWAS by identifying population-specific SNPs, showing its superiority over MuGLasso in simulations and real data.

The paper introduces AUC hyper-stability scores to improve the selection of Random Forest implementations for stable short biomarker signature discovery.

The paper introduces Komet, a scalable DTI prediction method, and LCIdb, a large DTI dataset, showing improved scalability and performance over existing methods.

pyComBat provides a Python implementation of ComBat and ComBat-Seq for batch effect correction in microarray and RNA-Seq data, offering similar performance with improved speed and open-source accessibility.

The paper evaluates methods to detect genes with differential expression dispersion in cancer datasets, identifying autophagy-related functions associated with tumor progression.

This study identifies gene-gene interactions in multiple sclerosis using GWAS data, finding 4 missense variant pairs and 117 eQTL-mediated pairs, including known interactions like GLI-I and SUFU binding.

The paper develops a network-guided method for detecting gene-level epistasis in GWAS data, showing it can identify disease-relevant interactions while controlling type I error.

This paper introduces a multitask group Lasso method for GWAS in diverse populations, improving feature selection and handling population structure, and shows it outperforms existing methods on simulated and real cancer data.

This paper proposes a method to reduce false positive drug target predictions by balancing negative examples in training data, improving prediction accuracy and reducing experimental costs.

The paper compares six network methods using GWAS data from the GENESIS study to identify breast cancer susceptibility genes, finding that network approaches improve results over standard GWAS.

Martini uses SNP networks to enhance GWAS by identifying functionally related SNPs, improving discovery and interpretability.

This paper introduces epiGWAS, a method for detecting epistasis in genome-wide association studies, showing improved identification of gene-gene interactions compared to existing approaches.

This paper introduces kernelPSI, a method for nonlinear post-selection inference in GWAS, showing it improves statistical power and interpretability compared to existing tools.

The paper introduces block HSIC Lasso, a nonlinear feature selection method for ultra-high dimensional data, showing it retains more biological information than existing techniques in synthetic and real genomic data.

The paper develops NN-MT, a multi-task SVM algorithm for predicting drug-protein interactions, showing efficient performance in proteome-wide and orphan case predictions.