Juan C. Caicedo

mnDINO is a vision transformer model for accurate micronuclei segmentation in DNA-stained images, achieving high accuracy and generalization across cell lines and microscopes.

The paper uses image-based profiling of hiPSC-derived trophoblasts to identify SCZ risk signatures linked to sex, placental genomic risk, and hypoxic stress.

This paper introduces TensionDL, a deep learning model that predicts pN receptor tension maps from cell morphology and vinculin images, validated across cell types and substrate stiffnesses.

The paper introduces a dataset of cell morphology images treated with matched chemical and genetic perturbations and evaluates methods for identifying perturbation similarities.

Self-supervised DINO features, learned without labels, describe cell morphology from subcellular detail to experimental groups and expose a hierarchy of biological and technical variation in three public datasets.

The paper predicts compound activity using chemical structures, images, and gene-expression profiles, showing that combining these modalities can improve prediction accuracy in drug discovery.

This study evaluates deep learning strategies for nucleus segmentation in fluorescence images, showing U-Net outperforms classical methods and reduces biologically relevant errors.

ImageNet-pretrained Inception-v3, without fine-tuning or segmentation, reaches 91% mechanism-of-action accuracy on illumination-corrected BBBC021 versus 90% for classical mean profiles, running in 22 minutes versus over 10 hours for single-core CellProfiler.