Fig. 1

CNRein uses evolutionarily aware deep reinforcement learning to perform single-cell copy number calling. a Technologies such as 10x Genomics CNV solution [26], DLP+ [27, 28], and ACT [29] enable high-throughput single-cell DNA sequencing suitable for CNA calling. b Reads are aligned and processed to obtain CNRein inputs. c CNRein uses a generative model of CNA evolution parameterized by a convolutional neural network and optimized via reinforcement learning. d From the evolutionary model \(\theta\), CNRein produces haplotype-specific copy number profiles P for each cell