#### ---- Description
This dataset contains single-cell RNA-seq, surface protein, and TCR sequencing data generated from T cells using the 10x Genomics 5' kit.
Both datasets were processed using Cell Ranger version 6.1.2 (10x Genomics).
A pretrained variational autoencoder (VAE) model is also provided for detecting CD4 CTLs.

#### ---- Revision history
2026-09-07
  Dataset1.tar.gz    unchanged
  Dataset2.tar.gz    UPDATED - filtered_feature_bc_matrix/ (barcodes, features, matrix)
                     replaced; the initial release contained Dataset1's files by mistake.
  VAE_model.tar.gz   unchanged

#### ---- Extract the archives
tar -zxvf Dataset1.tar.gz
tar -zxvf Dataset2.tar.gz
tar -zxvf VAE_model.tar.gz

#### ---- Dataset1.tar.gz
## T cells from 28 individuals (8 controls, 10 centenarians, 10 supercentenarians)
Dataset1/
├── cell.sample.cluster.43584.list (cell ID, sample ID, cluster name)
├── filtered_feature_bc_matrix/
│     ├── barcodes.tsv.gz
│     ├── features.tsv.gz
│     └── matrix.mtx.gz
├── vdj_t/
       ├── clonotypes.csv
       └── filtered_contig_annotations.csv


#### ---- Dataset2.tar.gz
## T cells from 6 individuals with and without PMA+ionomycin stimulation (2 controls, 2 centenarians, 2 supercentenarians)
Dataset2/
├── cell.sample.cluster.6400.list (cell ID, sample ID, cluster name)
├── filtered_feature_bc_matrix/
│     ├── barcodes.tsv.gz
│     ├── features.tsv.gz
│     └── matrix.mtx.gz
├── vdj_t/
       ├── clonotypes.csv
       └── filtered_contig_annotations.csv


#### ---- VAE_model.tar.gz
## Pre-trained VAE model and Python script for inference
VAE_model/
├── vae.py
├── 90.model/
│     ├── 01.gene.list
│     └── 02.model.best.weights.h5
└── 91.input/ (Test Data)
      ├── barcodes.tsv.gz
      ├── features.tsv.gz
      └── matrix.mtx.gz

#### ---- Example: Loading Dataset1 into Seurat
Seurat R package (version 4.1.1) in R (version 4.1.3)

## Load Seurat Package
library(Seurat)

## Define paths
dir  <- "Dataset1/filtered_feature_bc_matrix/"
cell <- "Dataset1/cell.sample.cluster.43584.list"

## Load RNA and protein data
data <- Read10X(data.dir = dir)

## Remove Hashtag antibodies (used for sample multiplexing; not used for clustering)
hashtags  <- c("H1_TotalSeqC", "H3_TotalSeqC", "H5_TotalSeqC")
data[["Antibody Capture"]] <- data[["Antibody Capture"]][!rownames(data[["Antibody Capture"]]) %in% hashtags, ]

## Create Seurat Object
tc              <- CreateSeuratObject(counts = data$`Gene Expression`)
tc[['Protein']] <- CreateAssayObject(counts = data$`Antibody Capture`)

## Load metadata and filter cells used in downstream analysis
tbl <- read.table("Dataset1/cell.sample.cluster.43584.list", row.names=1, col.names=c("bc","id","cluster"))
tc  <- subset(tc, cells = rownames(tbl))
tc  <- AddMetaData(tc, metadata = tbl)

#### ---- Example: Running VAE Inference
## Prerequisites: TensorFlow 2.20.0 or higher is required.
## (Using older versions will cause weight loading errors.)

## 1. Environment Setup
conda create -n vae python=3.10 -y
conda activate vae
pip install --upgrade pip
pip install "tensorflow[and-cuda]"
pip install numpy pandas scikit-learn scipy

## 2. Version Check
python -c "import tensorflow as tf; print(tf.__version__)"

## 3. Run Inference
cd VAE_model
python vae.py 91.input

## 4. Output Files
## Results are written to:
## 92.output/
## ├── 21.valus.infer.txt (log file)
## ├── 22.laten.infer.tsv (16-dimensional latent space)
## └── 31.label.infer.tsv (Cell type prediction results: CD4_CTL, non_T, other_T)