Processed input data for scFuseCpG single-cell methylome imputation benchmarks

This dataset contains the processed input files used for scFuseCpG, a single-cell methylome imputation framework based on graph-guided dense feature fusion. The archive includes processed benchmark and validation data prepared for model training, evaluation and downstream analysis. The processed files are organized by dataset and include scFuseCpG-compatible input files such as DNA sequence dictionaries, methylation matrix dictionaries, CpG genomic position dictionaries and associated metadata where applicable. These inputs were generated from public single-cell DNA methylation datasets and an external SeekGene_PBMC dual-omics validation dataset, following the preprocessing procedures described in the accompanying manuscript. The included benchmark datasets cover HCC, Hemato, Neuron-Mouse and Neuron-Homo single-cell methylome datasets, and the external validation dataset includes SeekGene_PBMC. Methylation states are represented in a binarized format, where observed CpG methylation states are encoded as methylated or unmethylated entries and unobserved cell–CpG entries are encoded as missing values. These processed inputs are provided to support reproducibility of the scFuseCpG experiments, including baseline comparison, model training, methylation-state prediction and downstream cell-type structure analysis.