Section 7 of 7
Supporting information
Guoxia Wen, Jiaqi Li, Hanyu Wu, Jialing Fang, Yuting Fu, Mengmeng Jiang, Yuqing Mei, Rui Xu, Yuxuan Du, Siran Chu, Guoji Guo, Xiaoping Han, and Jingjing Wang · about 1 minutes
Figure S1: Predictive performance of the CNN model on the different scRNA‐seq technologies.
Figure S2: Scatter plot of actual and predicted gene expression values from the logistic regression analysis trained on the mESCs dataset, Human PBMCs datasets and Pancreas dataset.
Figure S3: Benchmark of hyper‐parameters on predictive performance of the CNN model.
Figure S4: The exploration of advanced model architectures.
Figure S5: The exploration of Transformer architectures in the 10X genomics platform.
Figure S6: Function enrichment on genes from MCA embryonic dataset that predicted to expressed but not detected by scRNA‐seq.
Figure S7: Evaluation of the CNN model with MTL framework on rare cell types.
Table S1: Characteristics of different single‐cell sequencing methods used in the study.
Table S2: Performance details of scRNA‐seq methods for the CNN model.
Table S3: Statistical comparison of baseline and MAGIC‐imputed data across scRNA‐seq platforms.
Table S4: Summary statistics of normalized expression values across three datasets.