Large-scale proteomic inference at single-cell resolution by scInfer

Obtaining high-throughput, large-scale, and paired transcriptomic and proteomic data at the single-cell level is crucial for understanding the complex functions and phenotypic characteristics of multicellular organisms. However, current biomolecular measurement technologies are limited by their ability to detect only a small subset of functional proteins or by low cellular throughput, which hinders comprehensive analysis of cell function. Therefore, there is an urgent need for computational approaches that can bridge the gap between the high-throughput nature of single-cell RNA sequencing (scRNA-seq) and the large-scale protein profiling offered by single-cell proteomics. To address this challenge, we propose scInfer, a novel method that leverages single-cell proteomic data as a reference to infer large-scale protein expression profiles for each cell in scRNA-seq data. scInfer consists of two key modules: a self-supervised contrastive learning module that aligns unpaired transcriptomic and proteomic data, and an unsupervised weight generation module that performs the inference. We systematically evaluate the performance of scInfer on multiple datasets and demonstrate that the inferred protein expression closely matches experimentally measured values. scInfer enables effective downstream tasks such as differential protein identification and cell clustering. Moreover, it outperforms existing methods in multi-omics integration, significantly enhancing capabilities in cell subtype annotation, drug mechanism exploration, and the construction of single-cell multi-omics atlas.