Fig. 7

DeepTrace outperformed other models in identifying HCC-derived reads. The performance of DeepTrace in identifying HCC-derived reads was compared with that of LSTM, GRU, CNN + LSTM, and CNN + GRU based on the read datasets before and after noise reduction separately. The read dataset after noise reduction filtered out noise reads according to the rules described in the “Data noise reduction” method. Panel A shows the accuracy, f1 score, MCC of different models in identifying HCC-derived reads in the reads test datasets. Panel B shows the precision recall curve and panel C shows the ROC curve of different models