Abstract 906: Large-scale surface protein abundance prediction from single-cell transcriptomes with a zero-shot model and its applications to cancer research

Abstract

Abstract We develop SPIDER, a zero-shot model which can predict the abundance for a large scale (>2,000) of surface proteins from single-cell transcriptomes in various contexts. This is to overcome the challenges in current single-cell protein abundance quantification tools (e.g., flow cytometry, CITE-seq) and computational models (e.g., totalVI, Seurat, cTPnet), where routinely only <300 surface proteins can be quantified or predicted. Comprehensive benchmarking on four external validation sets shows that the prediction accuracy of SPIDER outperforms other state-of-the-art methods including Seurat, totalVI, and cTPnet, with a prediction accuracy elevated as much as 42.5% in new contexts. We further conduct case studies in cancers by applying SPIDER to predict the abundance of >2,500 surface proteins on scRNA-seq datasets of hepatocellular carcinoma (HCC) and colorectal cancer liver metastasis (RCRLM), where the predicted surface protein abundance data is analyzed to demonstrate the broad downstream applications of SPIDER including facilitating cell type annotation, disease biomarker/target identification, and cell-cell interaction (CCI) inference. For instance, in these datasets we find that SPIDER can reveal (novel) disease biomarkers that are overlooked by solely observing transcript expression, such as CD44/PIK3IP1 in hepatocytes/γδ2 T cells as a positive/negative HCC biomarker, demonstrating the capability of SPIDER to compensate for the limitations in the RNA modality. In conclusion, we propose the SPIDER model which can provide valuable information on the abundance of a large scale of surfaceomes in single cells, and can be further used to gain new insights into cancers via promoting cell type annotation, disease biomarker/target identification, and CCI inference. Citation Format: Ruoqiao Chen, Bin Chen, Jiayu Zhou. Large-scale surface protein abundance prediction from single-cell transcriptomes with a zero-shot model and its applications to cancer research [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 906.