Fig. 1
From: Inferring single-cell copy number profiles through cross-cell segmentation of read counts
The workflow of DeepCNA. To improve read counts obtained from scDNA-seq data, DeepCNA takes several data correction and normalization procedures including correcting GC-content and mappability bias, eliminating outlier bins with extremely high or low read counts, and removing outlier cells with low sequencing quality. Given the normalized read counts, DeepCNA employs an autoencoder network to learn low-dimensional latent representation of each genomic bin, such that the breakpoints can be accurately detected along each latent dimension and merged together to form the final breakpoints. Finally, a mixture model is adopted to estimate copy numbers of the inferred segments for each cell