3

scConcept: Contrastive pretraining for technology-agnostic single-cell representations beyond reconstruction
We present scConcept, a transformer-based contrastive learning framework for single-cell transcriptomics that learns technology-agnostic cell embeddings by contrasting multiple views of cells, outperforming foundation models and domain-specific methods across diverse downstream tasks.
Mojtaba Bahrami, A Tejada-Lapuerta, S Becker, FS Hashemi G, FJ Theis
PDF Cite Code DOI
Optimal distance metrics for single-cell RNA-seq populations
We benchmark 16 distance metrics for single-cell RNA-seq and recommend mean squared error (MSE) for comparing populations and evaluating gene expression prediction models.
Y Ji, TD Green, S Peidli, Mojtaba Bahrami, M Liu, L Zappia, K Hrovatin, C Sander, FJ Theis
PDF Cite Code DOI