scMARK an 'MNIST' like benchmark to evaluate and optimize models for unifying scRNA data

<p>Here we present a novel benchmark dataset (scMARK.v2), that consists of 11 published cancer scRNA-seq studies, for which&nbsp;we standardized&nbsp; cell-type author labels and gene identifiers. scMARK.v2 can be used to ask how well models integrate data from different scRNA studies. We also provide&nbsp;a 12th standardized study (Wu&nbsp;et al 2021) that we&nbsp;held-out for evaluation of alignment of data &quot;never seen&quot; before, and a 13th study of newly generated in-vitro scRNA-seq data from cancer and fibroblast cells.</p> <ul> <li>Data is provided as aData *h5ad files that can be read with Python&#39;s library <a href="https://scanpy.readthedocs.io/en/stable/">Scanpy</a>.</li> <li>Studies inclided in scMARK.v2 were downsampled to 10,000 cells per study.</li> <li>The difference between&nbsp;<a href="https://zenodo.org/record/5765804">scMARK.v1</a> and&nbsp;scMARK.v2, is that in v2,&nbsp;we provide at least two studies for each cancer type and each cell type; whereas in v1&nbsp;a handfull of cell types were present only in one study.</li> </ul>