anasanchezf/plate_effects
Multi-Source Domain Adaptation for Bioimaging Data (MSCDA-BioIm) MSCDA-BioIm is a biomedical microscopy benchmark for evaluating test-time and in-context domain adaptation under realistic batch effects. Built from the large-scale JUMP-CP dataset, it targets mechanism-of-action (MoA) classification using five-channel images of compounds associated with eight well-defined MoA classes. The dataset is organized by experimental batches and imaging… See the full description on the dataset page: https://huggingface.co/datasets/anasanchezf/plate_effects.
<p align="center"> <a href="https://arxiv.org/abs/2604.20824"> <img src="https://img.shields.io/badge/arXiv-2604.20824-b31b1b.svg" alt="arXiv"> </a> <a href="https://github.com/ml-jku/cs-arm-bn"> <img src="https://img.shields.io/badge/Github-181717?logo=github&logoColor=white" alt="GitHub"> </a> </p>
<p align="center"> <img src="figures/data_overview.png" alt="Dataset overview" width="750"/> </p>
<p align="center"> </p>
Multi-Source Domain Adaptation for Bioimaging Data (MSCDA-BioIm)
MSCDA-BioIm is a biomedical microscopy benchmark for evaluating test-time and in-context domain adaptation under realistic batch effects. Built from the large-scale JUMP-CP dataset, it targets mechanism-of-action (MoA) classification using five-channel images of compounds associated with eight well-defined MoA classes. The dataset is organized by experimental batches and imaging sources, enabling controlled evaluation of generalization to unseen batches, cross-source transfer, small target context sizes, and label-shifted target batches.
A key feature of MSCDA-BioIm is the inclusion of negative control samples in every experimental batch. These unperturbed samples provide a stable reference for estimating batch-specific technical variation, making the dataset especially suitable for studying control-aware adaptation methods such as CS-ARM-BN.