Statistical Methods and Inference

This paper introduces sparse latent factor regression models for genome-wide and epigenome-wide association studies, showing improved statistical performance and robustness in both simulated and real data applications.

SurvBenchmark evaluates survival analysis methods on clinical and omics data, showing model performance varies with metrics and datasets.

This paper introduces Xsurv, an R package using XGB and LGB for survival analysis, showing it outperforms traditional methods in biomarker discovery for cancer prognosis.

The paper proposes a stationary Gaussian graphical model for microbial interaction networks using longitudinal 16S rRNA data and shows it outperforms conventional methods in correlated settings.

The paper introduces penalized generalized estimating equations for relative risk regression to address instability in modeling brain lesion data, showing improved convergence and interpretation.

This paper introduces a penalized linear mixed model with generalized method of moments for predicting complex phenotypes from genomic data, showing improved performance over existing models.

The paper introduces a hierarchical resampling method for multi-study prediction, showing improved performance in neurochemical sensing applications.

ParProx enables scalable multivariable regression analysis of ultrahigh-dimensional omics data using group lasso with biological priors, producing interpretable models.

The paper develops efficient solvers for large-scale regularized regressions on genetic data, reducing memory usage and improving computational speed.

The paper develops MAGIC-LASSO to predict missing data in biobanks with non-random missingness, achieving high phenotypic and genetic correlations in UK Biobank AUDIT score predictions.

This paper introduces a dual form of the Cox partial likelihood loss using Fenchel duality, enabling efficient kernel learning for survival analysis with censored outcomes.

The paper introduces a smoothed Lasso method for high-dimensional regression and applies it to compute polygenic risk scores for COPD, showing comparable accuracy to FISTA with similar runtime.

The paper introduces a method for selecting Gaussian graphical models using prior knowledge and applies it to metabolomic data, finding that metabolic pathways significantly influence metabolite dependencies.

The paper introduces bFMD, a module detection algorithm that identifies balanced functional modules in genomic data, showing favorable performance in simulations and interpretable results in RNA-seq data.

This paper benchmarks the performance of two R packages, snpnet and bigstatsr, for fitting penalized regressions on large genetic data, finding bigstatsr significantly faster.

The paper improves random forest's power for high-dimensional microbiome data by combining forward variable selection with random forest, achieving higher accuracy in case-control studies.

This paper introduces a scalable framework for sparse regression on large-scale data, demonstrating improved prediction performance on UK Biobank traits using the BASIL method and snpnet package.

This paper introduces a smoothing framework for L1 penalized regression that provides accurate predictions with theoretical guarantees and comparable runtime to existing methods.

The paper introduces an R package implementing lasso, group lasso, and sparse-group lasso regularization via proximal gradient descent for high-dimensional linear regression.

The paper introduces a penalized regression method that integrates external meta-features a-priori to improve prediction and interpretation in genomic studies.

The paper introduces a group-regularized Cox regression method for survival analysis of rare events using common responses, showing improved prediction performance on UK Biobank data.

The paper introduces a compositional knockoff filter to control false discovery rate in high-dimensional microbiome regression analysis, demonstrating improved power while maintaining FDR control.

SuSiE writes sparse regression coefficients as a sum of single-effect vectors; its IBSS algorithm returns a credible set per effect and outperformed existing fine-mapping methods in numerical experiments.

The paper develops a scalable sparse reduced-rank regression method for high-dimensional multitask learning, applied to UK Biobank data to generate polygenic risk scores.

The paper introduces a statistical method to test for association between a response variable and a tree structure across all hierarchy levels while controlling false positives.

The paper introduces CALF, a linear regression method that outperforms LASSO in several metrics on real medical data.

This paper introduces Robocov, a convex optimization framework for estimating gene-level tissue networks with missing data, showing it improves correlation and causal structure estimation and identifies disease-related genes.

The paper introduces a method for integrating heterogeneous biological data using a graphical LASSO framework and applies it to study asthma and eczema risk factors in children.

The paper introduces methods for feature selection in genomics data with censored survival outcomes, addressing non-proportional hazards and outperforming existing approaches.

binomialRF identifies biomarkers and their interactions using a scalable random forest approach with statistical hypothesis testing for feature selection.