DECANT prioritises biological heterogeneity over batch effects for various single-cell omics data

Résumé fourni par la source

As single-cell transcriptomic and epigenomic atlases expand across samples, protocols, and laboratories, batch-associated variation increasingly obscures biological heterogeneity. While downstream integration partially mitigates this variation, the upstream trade-off between biological discrimination and batch association for individual features remains poorly explored. Here we introduce DECANT, an annotation-free framework that performs continuous joint prioritisation of biologically discriminative variation and batch association without hard feature exclusion or predefined batch penalties. We evaluated DECANT on nine datasets spanning single-cell transcriptomics, DNA methylation and chromatin accessibility, encompassing batch variation associated with technology, donor, brain region, condition and species. Across four modality-matched workflows applied to each dataset, DECANT systematically improved downstream performance in both biological conservation and batch mixing. Importantly, DECANT retained coherent gene programmes, preserved local structure in rare cell populations, supported cell-type-resolved regulatory interpretation, and revealed stronger tissue specificity in cortical methylation data. Such continuous prioritisation therefore provides effective upstream batch control across single-cell omics while preserving biological signals essential for downstream interpretation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.