Identification Of Platelets/Megakaryocytes In Single Cell Transcriptomic Data With The Assistance Of Machine Learning

Platelets play a crucial role in blood clotting and wound healing,they also serve as a key source of vascular endothelial growth factor and can regulate tumor angiogenesis and vascular integrity.They are closely associated with tumor immune escape and metastasis,playing an indispensable role in tumor development.However,due to their small size and lack of a nucleus,platelets are difficult to identify in single-cell transcriptome sequencing(sc RNA-seq)data,and they are prone to being overlooked or misidentified as noise in analysis.Therefore,it is particularly important to accurately identify platelets in sc RNA-seq data to explore the mechanisms related to the tumor microenvironment and development.In this study,a platelet annotation model,Pltscanner,was constructed by combining bulk and single-cell RNA sequencing data.Feature process selection was performed on the bulk data using a method based on the relative rank relationship of gene expression.An XGBoost algorithm was employed to build an integrated model on the single-cell RNA data,and parameter optimization was conducted.The model was evaluated using performance metrics obtained from the validation set and was also validated from three dimensions: the expression of platelet marker genes,clustering analysis with self-tested platelet data,and spatial transcriptomics vascular distribution.Finally,we developed an R package for the Pltscanner model and made it available on Git Hub(https://github.com/Ziru Huang/Pltscanner).The results showed that the Pltscanner model had a sensitivity of 99.8%,specificity of 97.1%,and an AUC of 0.975 on the validation set.The expression of platelet-specific genes in the platelets identified by Pltscanner was significantly higher than in other cells(p < 0.001),and they could be clustered with self-tested platelet data.In spatial transcriptomic data,the high predicted score regions overlapped with the vascular distribution area.Compared to existing cell annotation tools,Single R and Garnett,Pltscanner showed faster computation speed and better annotation performance.Additionally,this study analyzed the Pltscanner annotation results and found that platelets are not a single type and can be classified into lymphocyte-associated platelets(LAPs),macrophage-associated platelets(MAPs),and traditional platelets(TPs)based on different marker gene expression characteristics.The diversity of platelets suggests different roles and functions.Cell communication analysis revealed that in the tumor microenvironment of colorectal cancer,LAPs primarily affect tumor progression through the PI3K-Akt signaling pathway,while MAPs interact with tumor cells through the estrogen signaling pathway.The Pltscanner model efficiently and accurately identifies platelets in sc RNA-seq data,offering valuable insights into the altered gene expression patterns of platelets in the tumor microenvironment and revealing the underlying biological mechanisms of their interaction with tumor cells.This model lays a foundation for deeper exploration of the mechanisms behind tumorigenesis.