The 'LDA' algorithm is a creative and effective method to handle large scATAC dimensions. Unlike scRNA, which often genes x cells, ATAC peaks are hundreds of times more than gene numbers. However, in my own analysis experience, I got separative clusters (and further different developmental trajectories) of cells, which were often different with sequence-depth. I wonder that : 1) does the cisTopic LDA algorithm treat low-depth cells properly? will it possible that these clusters were only separated due to sequence depth? 2) should I filter these low-depth (in fact, not so low-depth at my first glance)
Please help and any suggestions will be good. I just want to keep off false-positive results but do not to miss the true findings.
p.s. the results of signac(seurat extension package), episcanpy & scanpy, cicero look similar.
The 'LDA' algorithm is a creative and effective method to handle large scATAC dimensions. Unlike scRNA, which often genes x cells, ATAC peaks are hundreds of times more than gene numbers. However, in my own analysis experience, I got separative clusters (and further different developmental trajectories) of cells, which were often different with sequence-depth. I wonder that : 1) does the cisTopic LDA algorithm treat low-depth cells properly? will it possible that these clusters were only separated due to sequence depth? 2) should I filter these low-depth (in fact, not so low-depth at my first glance)
Please help and any suggestions will be good. I just want to keep off false-positive results but do not to miss the true findings.
p.s. the results of signac(seurat extension package), episcanpy & scanpy, cicero look similar.