adverSCarial: a toolkit for exposing classifier vulnerabilities in single-cell transcriptomics
Résumé
Adversarial attacks pose a significant risk to machine learning (ML) tools designed for classifying single-cell RNA-sequencing (scRNA-seq) data, with potential implications for biomedical research and future clinical applications. We present adverSCarial, a novel R package that evaluates the vulnerability of scRNA-seq classifiers to various adversarial perturbations, ranging from barely detectable, subtle changes in gene expression to large- scale modifications. We demonstrate how five representative classifiers spanning marker- based, hierarchical, support vector machine, random forest, and neural network algorithms, respond to these attacks on four hallmarks scRNA-seq datasets. Our findings reveal that all classifiers eventually fail under different amplitudes of perturbations, which depend on the ML algorithm they are based on and on the nature of the modifications. Beyond security concerns, adversarial attacks help uncover the inner decision-making mechanisms of the classifiers. The various attack modes and customizable parameters proposed in adverSCarial are useful to identify which gene or set of genes is crucial for correct classification and to highlight the genes that can be substantially altered without detection. These functionalities are critical for the development of more robust and interpretable models, a step toward integrating scRNA-seq classifiers into routine research and clinical workflows. The R package is freely available on Bioconductor (10.18129/B9.bioc.adverSCarial) and helps evaluate scRNA-seq-based ML models vulnerabilities in a computationally-cheap and time-efficient framework.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |

