Extended Data Fig. 1: Evaluation of DeepTFni on human PBMC scATAC-seq dataset. | Nature Machine Intelligence

Extended Data Fig. 1: Evaluation of DeepTFni on human PBMC scATAC-seq dataset.

From: Inferring transcription factor regulatory networks from single-cell ATAC-seq data based on graph neural networks

Extended Data Fig. 1

(a). DeepTFni achieves over 0.84 accuracy across different cell types in human PBMC dataset. TF interaction numbers in initial adjacency matrix and prediction matrix are listed. (b). DeepTFni is well performed on the large scATAC-seq dataset, which contains ~800,000 cells from 15 human fetal organs. Blue bar indicates the cell number of each organ. Green bar indicates the number of ATAC peaks filtered in DeepTFni. Yellow bar indicates the running time of DeepTFni for each organ. (c). Jaccard index of interactions before and after masked-positive disturbance. Red line indicates the Jaccard index of DeepTFni prediction results on test set. Blue line indicates the Jaccard index of disturbed train and validate set. (d). Accuracy of DeepTFni prediction in disturbed dataset with different masked-positives proportion. The dashed line represents the accuracy without disturbance. (e). Number of false negatives in disturbed dataset with different masked-positive proportion. (f). Recovered ratios of masked-positives in DeepTFni prediction results. Black dots represents 5 times dataset disturbance.

Back to article page