The scRNA-seq data were collected from previously published datasets (GSE140228, GSE139555, GSE155698, GSE121636, and GSE139324), adhering to the following selection criteria: 1) presence of T cells, 2) treatment-naïve patients, 3) solid tumors, and 4) inclusion of at least tumor and blood samples.Each scRNA-seq dataset underwent separate preprocessing in R (v4.0.2). We filtered out cells from the original count matrices that had fewer than 200 genes detected or more than 10% mitochondrial UMI counts and we only kept genes detected in at least 3 cells. Then, we applied Seurat (v4.0.5) with default parameters for count data normalization and scaling. Each cell was assigned a cell cycle score using the CellCycleScoring function and we computed the difference between the G2M and S phase scores. This approach allows for the separation of non-cycling from cycling cells while minimizing the differences in cell cycle phase among proliferating cells. The SelectIntegrationFeatures function was ran with the nfeatures parameter set to 3,000 before merging all samples from each dataset. These integration features were then used for Principal Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP). Clustering was performed using the Louvain algorithm with the resolution parameter set to 2.0 for all datasets. Finally, T cells were isolated based on CD3D and CD3G genes expression (CD3D or CD3G expression level > 0). To integrate heterogeneous data from different sources, a two-step procedure was applied. We first concatenated all datasets together and ran the scaling and PCA steps based on the top 3,000 highly variable genes identified by the FindVariableFeatures function with the “vst” method. Harmony was applied for batch effect correction then UMAP and clustering using the Louvain algorithm with the resolution parameter set to 2.0 were performed on the harmony reduction. Examining the result from the first clustering run, we identified contamination clusters and clusters that arose from unwanted factors: we removed the contamination clusters including low quality cells highly expressing marker genes associated with apoptosis and tissue dissociation operation, pancreatic acinar cells (expressing PRSS1, CLPS, PNLIP and CTRB1 among others), myeloid cells (expressing CD68) and B cells (expressing CD79A). Then, we performed the second run of integration and clustering excluding immunoglobulin, ribosome-protein-coding, and T cell receptor (TCR) genes (gene symbol with string pattern "^IGK|^IGH|^IGL|^IGJ|^IGS|^IGD|IGFN1", "^RP([0–9]+-|[LS])", and "^TRA|^TRB|^TRG" respectively) from the top 3,000 highly variable genes and regressing out the cell cycle difference effect as well as the percentage of mitochondrial UMI counts. Harmony (v0.1.0) was applied again for batch effect correction and UMAP was performed on the harmony reduction.T cell subtypes identification and annotation was performed by clustering cells using the Louvain algorithm with the resolution parameter set to 4.1 after iterative testing from 3.5 to 5.0 by 0.1 (more granular than default), computing clusters signatures based on differential gene expression using the FindAllMarkers function with the “MAST” method and interrogating known gene markers expression. A resolution value of 4.1 was notably found to be the lowest resolution value enabling the correct separation of proliferating CD4+ T cells from proliferating CD8+ T cells.