Genes with high column norms in this matrix are the top attributed genes per latent dimension, representing learned gene programs. Marker overlap (MO) is computed as the fraction of differentially ...