Pedro J. Ballester

This study evaluates Conformal Prediction for predicting cancer cell growth inhibition by novel molecules, showing that smaller confidence intervals correlate with lower prediction errors across diverse cell lines.

This study evaluates transfer learning strategies for predicting chemotherapy response using clinical data, finding that hybrid and fine-tuning approaches perform better than direct model transfer or biomarker-based methods.

This study uses machine learning on multi-omics data to predict temozolomide response in low-grade glioma patients, finding that miRNA-based models outperform conventional biomarkers.

SCORCH2 improves virtual screening performance and interpretability by using interaction features, showing better accuracy and transferability than its predecessor.

The paper develops a pan-cancer, pan-treatment model to predict drug responses using patient-derived xenografts, showing it works well across cancer types with limited training data.

The paper uses machine learning to predict drug responses in colorectal cancer cells, identifying novel compounds with cytotoxic or cytostatic effects.

The paper identifies DNA methylation subtypes from transcriptomes to guide sunitinib and immunotherapy selection in renal cancer, finding hyper-methylated tumors respond better to atezolizumab plus bevacizumab.

The paper evaluates machine learning algorithms for phenotypic virtual screening of cancer cell lines and finds that a graph-based deep neural network performs best.

The paper develops and evaluates machine learning scoring functions for identifying PARP1 inhibitors, finding a support vector machine-based model with PLEC fingerprints outperforms others in virtual screening.

This study shows that larger compound libraries improve the performance of scoring functions in structure-based virtual screening across multiple targets and docking tools.

The paper predicts synergistic cancer drug combinations using NCI-ALMANAC data with machine learning models, achieving high accuracy and reliability.

The paper presents a large-scale cancer pharmacogenomic screen with 11,576 experiments on 910 drug combinations across 85 cancer cell lines, enabling crowd-sourced prediction of synergistic drug pairs and biomarkers.

The paper uses machine learning to predict drug response in tumour models by reducing molecular profile dimensionality, finding multi-gene predictors outperform single-gene markers in several cancer types.

The paper introduces a non-parametric method to measure discriminative power of drug-gene associations and identifies 128 new genomic markers across 61% of analyzed drugs.

CCLP predicts compound activity across the NCI60 panel using a multi-task Random Forest model and provides prediction errors via conformal prediction.

The paper compares single-gene and multi-gene predictors of drug sensitivity in cancer cell lines using gene expression data and finds that multi-gene models perform better overall, though single-gene markers have higher precision.