Ewa Szczurek

The paper introduces FACTMx, a variational framework that jointly models structured and simple patient-level data to learn interpretable representations, showing improved performance in simulations and applications to cancer and cardiovascular cohorts.

OmegAMP generates antimicrobial peptides with controlled physicochemical properties and safety profiles, showing activity against drug-resistant bacteria in vitro and in vivo.

BATTLE-AMP benchmarks antimicrobial peptide predictors using MIC data, showing that models trained on MIC data outperform binary classifiers and highlighting limitations in distinguishing active from inactive peptides.

This study identifies distinct pneumonia endotypes based on pathogen-specific host responses in the lung, linking them to clinical outcomes and suggesting personalized treatment strategies.

This paper introduces GUT-FORMer, a transformer-based autoencoder for gut microbiome analysis, achieving accurate disease classification and host trait prediction.

The paper integrates kidney and urine proteomics with peptidomics and machine learning to identify encrypted antimicrobial peptides as functional effectors in human pyelonephritis.

CLING is a multi-view factor model that improves factor and loading recovery in multi-omic data by using nonparametric Gamma shrinkage, outperforming baselines on synthetic and real datasets.

Cellohood identifies cellular neighborhoods in spatial transcriptomics data, revealing clinically relevant patterns across multiple granularity levels.

PepGlider enables precise control over peptide properties through a continuous regularization framework, allowing for de novo generation and targeted optimization of peptides with desired characteristics.

The paper presents two workflows for analyzing multiplexed imaging data from cancer tissues, applied to over 10,000 samples from 2500 patients, enabling large-scale tumor microenvironment studies.

DelSIEVE infers cell phylogeny and variants from scDNA-seq data, improving detection of deletions and evolutionary events compared to existing methods.

CaClust integrates exomic, scRNA-seq, and BCR data to infer clone genotypes and cell-to-clone mappings in follicular lymphoma, revealing genotype-phenotype links.

This paper introduces ST-Assign, a Bayesian model that jointly annotates cell types in scRNA-seq and decomposes cell-type mixtures in spatial transcriptomics data, showing improved accuracy over existing methods.

Celloscope deconvolves cell types in spatial transcriptomics data using marker genes, outperforming existing methods and accurately identifying brain structures and immune infiltrates.

This paper introduces HydrAMP, a deep generative model for creating potent antimicrobial peptides, which outperforms existing methods and validates nine new peptides in lab tests.

SIEVE jointly infers somatic variants and cell phylogeny from scDNA-seq data, showing improved accuracy in both tasks compared to existing methods.

This study uses deep learning to segment tumor microenvironments in lung cancer images, predicting patient survival and gene mutations with moderate to high accuracy.

SLIDE-VIP identifies synthetic lethal gene pairs in DNA damage repair, chromatin remodeling, and cell cycle using multi-omics data and eight statistical tests, reducing candidate pairs by 250-fold.

CONET infers tumor evolutionary trees and copy number calls from single-cell data using a probabilistic model with MCMC and regularization.

The paper develops a deep learning model called DEERS to predict kinase inhibitor efficacy in cancer cell lines, achieving high accuracy and providing interpretable insights into drug response mechanisms.

The paper develops a model to infer pairwise epistatic gene interactions from siRNA perturbation screens, showing that glinternet outperforms xyz in accuracy for high-dimensional data.

The paper evaluates feature selection strategies for drug sensitivity prediction, finding that prior knowledge-based methods outperform genome-wide approaches for drugs targeting specific genes and pathways.

The paper introduces a Bayesian deep learning framework for histopathological image classification, achieving high accuracy and providing uncertainty measures to improve active learning efficiency.

This paper introduces SurvLRT, a statistical method to identify epistatic gene pairs and triplets from cancer genomic and survival data, demonstrating its utility in finding biomarkers for synthetic lethal therapy.