Book This lecture, Professor Chen Shengquan, focused on common problems and technological breakthroughs in the field of single cell epigenomic data analysis. It is pointed out that there are significant shortcomings in current single-cell epigenetic data. Not only do they have data characteristics of millions of ultra-high dimensions, more than 90% of values ​​are missing, and extremely low signal-to-noise ratio, they also face multiple computational bottlenecks such as batch effect interference, difficulty in identifying rare cells, high computational pressure on tens of millions of cells, and false negative bias, which greatly limits the advancement of related research. In response to the above-mentioned industry pain points, Professor Chen Shengquan's team independently developed a three-layer integrated algorithm framework of unsupervised, weakly supervised and supervised to break through technical barriers in all aspects. Among them, the unsupervised system relies on adaptive iterative optimization to complete core processing such as data noise reduction and batch effect correction. The innovative mapping scheme takes into account bias elimination and preservation of real biological differences, significantly reducing the cost of map reconstruction. The weakly supervised model integrates single cell and spatial reference data, uses the Bayesian graph model to mine spatial correlations between cells, and revitalizes public data resources. The supervised framework relies on contrastive learning and graph networks to achieve intelligent annotation of cells and identify new subtypes. It also reveals for the first time security risks such as backdoor attacks and data poisoning in large single-cell models, providing a new basis for data quality control. The interactive session of the lecture was lively and the participants had in-depth exchanges on practical difficulties such as cross-omics integration, very large data set operations, cross-species data fusion, and model optimization. Finally, Professor Chen Shengquan pointed out that at present, the relevant algorithms have only completed theoretical verification. In the future, cross-border collaboration with biological laboratories will be deepened to customize optimization models for core areas such as tumors, nerves, and embryonic development, promote algorithm implementation and wet experimental verification, and truly realize data-driven innovative research in life sciences.