Given a reduced and denoised dataset, we next sought to learn the shape of the manifold of cells (that is, the underlying lower-dimensional gene expression space on which cells are preferentially located). Examining the PCA revealed that the manifold consisted of feather-like, elongated structures, extending variously into the different principal components. We found that the manifold was structured at many levels, ranging from broadly different classes of cells, individual cell types, to more subtle sub-types or states.
