Computational Methods for Mass Spectrometry-Based Single-Cell Proteomics Data

Abstract

Mass spectrometry-based single-cell proteomics (MS-SCP) enables the quantification of protein abundance in individual cells, offering a molecular perspective on post-transcriptional regulation and heterogeneity that cannot be inferred from transcriptomic data alone. However, MS-SCP data exhibit high rates of missing values, batch effects, and low-input noise, which require tailored computational models. In this review, we examine computational developments across the MS-SCP pipeline, covering protein identification and quantification, public repositories, data enhancement, and downstream analysis. We emphasize algorithms that integrate statistical modeling and deep learning for identification, quantification, and the joint correction of missingness and batch effects. We highlight the unique role of deep learning in modeling non-linear batch-dependent effects and learning robust protein representations from sparse, high-dimensional MS-SCP data. Finally, we outline future directions for method developers, including the incorporation of biological priors, the construction of abundance-level foundation models, the curation of single-cell perturbation datasets, and the integration of proteomic information with spatial and multimodal data.