We have single cell RNA-seq data and spatial transcripton data from 10x spatial transcriptomics technique to decoding the tumor cells heterogeniety. We are going to use Cell2location to deciphering the cell type population in each spatial location in each sample. We have sequenced 30 tumor samples using the 10x spatial transcriptomics technique to explore tumor cell heterogeneity. To analyze these samples, we plan to use Cell2location, a tool designed to identify cell type populations in each spatial location within each sample. Since Cell2location relies on machine learning algorithms implemented in PyTorch, it requires GPU resources for efficient processing. The Cell2location software involves the following key steps: Simulating the abundance of cell types across spatial locations. Simulating the expected multi-cell mRNA expression of genes across locations. Simulating multi-cell mRNA integer counts, accounting for differences in technology effects. During the training process, Cell2location utilizes hyperparameters, with each simulation requiring 30,000 iterations. Each simulation is expected to take approximately 10 hour. Given that we have 30 samples to analyze, and plan to run each dataset three times to ensure reproducibility, we anticipate a total of 100 simulations, amounting to 1000 GPU hours.