LucaCell: a sequence-centric foundation model for cross-species single-cell analysis

Single-cell foundation models have transformed transcriptomic analysis, yet most rely on fixed gene identifiers that limit transfer across species and data types. Here we present LucaCell, a sequence-centric foundation model that represents genes through pre-trained mRNA sequence embeddings rather than static gene annotations. Gene expression is discretized into bins and modeled with a Transformer encoder, enabling sequence-informed cell representations that generalize across species and data types. Pre-training on 85 million human and mouse single cells, LucaCell enables cross-species and cross-modal cell type transfer in human, mouse, and lemur kidney datasets; alignment-free microbial embedding that distinguishes more than 50 bacterial species while preserving intra-species physiological states; and accurate host-virus interaction prediction across five influenza A virus strains. In addition, LucaCell incorporates donor-specific exonic SNPs to refine gene expression reconstruction and improves perturbation response prediction. Together, these results show that transcript-sequence information provides a transferable prior for single-cell foundation modeling across species, modalities, and biological contexts.