The summary of the tasks is as follows.

1. First, you need to develop a computational method that learns 10-dimensional latent space representation for both the scRNA-seq dataset as well as the Protein dataset, resulting in output with 20000*10 dimensions.
2. Second, you need to perform clustering on the latent space representation using algorithm of your choice (leiden, louvain, kmeans, etc). Please note that the number of clusters in the dataset is 8. In case you are using the Louvain or Leiden algorithm, you should use a resolution that results in 8 clusters. The clustering output format is described in the assignment pdf.
3. Third, you need to plot the latent space representation using UMAP (Uniform Manifold Approximation and Projection). UMAP is primarily a dimensionality reduction technique that is commonly used to visualize high-dimensional data by reducing it to 2D or 3D. The UMAP algorithm are available in Scanpy package, Python https://scanpy.readthedocs.io/en/stable/api/generated/scanpy.pl.umap.html, Seurat package, R https://satijalab.org/seurat/reference/runumap. Helpful tutorials : https://scanpy.readthedocs.io/en/stable/tutorials.html, https://satijalab.org/seurat/

Lastly, you will be able to calculate ARI as a measure of your clustering accuracy by submitting to Kaggle.