Integrate and generate single-cell proteomics from transcriptomics with cross-attention

Abstract Motivation Single-cell RNA sequencing (scRNA-seq) and cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) have experienced rapid advancements in recent years, accompanied by the development of numerous methods for analyzing scRNA-seq and CITE-seq data. These innovations have enabled deeper insights into cellular heterogeneity and functional phenotypes. However, analyzing scRNA-seq and CITE-seq data within a unified framework remains a significant challenge in the field of single-cell analysis. Specifically, this challenge centers on two primary objectives: aligning scRNA-seq and CITE-seq cells within an integrated representation space and generating antibody-derived tag (ADT) measurements for scRNA-seq cells. Results By incorporating interrelationships between cells into a deep generative model with cross-attention, we introduced scProca to integrate and generate single-cell proteomics from transcriptomics. scProca delivers state-of-the-art performance in both integration and generation tasks across benchmark datasets. Furthermore, scProca can accommodate cells across experimental batches, showcasing its flexibility in complex experimental contexts. Availability The code of scProca is available at https://github.com/xiongbiolab/scProca , and replication for this study is available at https://github.com/ZzzsHuqiaAao/scProca-reproducibility .