Martin Ester

scMUSCL improves scRNA-seq clustering by transferring knowledge from multiple annotated datasets without requiring batch correction or prior cluster number knowledge.

ScRAT predicts clinical phenotypes from scRNA-seq data using attention-based neural networks, outperforming existing methods with limited samples and identifying critical cell types.

The paper develops a deep learning model to predict ligand-target interactions using protein structure graphs and attention mechanisms, achieving state-of-the-art performance on benchmark datasets.

This paper provides guidelines for developing machine learning models to predict drug sensitivity using cell line-based pharmacogenomics data, emphasizing proper training and validation practices.

Velodrome improves drug response prediction by generalizing across data distributions using labeled and unlabeled gene expression data from diverse sources.

DESMOND identifies differentially expressed gene modules in heterogeneous diseases by integrating gene expression and protein interaction data, improving reproducibility and biological relevance.

AITL improves drug response prediction by addressing input and output space discrepancies between pre-clinical and clinical pharmacogenomics datasets.

This paper introduces BDKANN, a neural network using biological knowledge to predict cancer drug response from gene expression data, achieving low error and interpretable results.

This paper introduces a mathematical model to identify subtype-specific temporal order of cancer pathway dysregulation and demonstrates its effectiveness on glioblastoma and colorectal adenocarcinoma data.

MOLI integrates multi-omics data using deep neural networks to predict drug response, achieving higher accuracy than single-omics and early integration methods.

This study introduces DGS, a method combining denoising autoencoders and transfer learning to predict prostate cancer metastasis using both labelled and unlabelled gene expression data, achieving higher accuracy than existing signatures.

SUBSTRA is a Bayesian method for patient stratification that balances predictive performance and interpretability by using regularized biclustering and iterative feature weighting.