Fig. 9
From: Identifying piRNA targets on mRNAs in C. elegans using a deep multi-head attention network
The devised deep learning network is robust against different extended mRNA lengths (l). We trained the devised model architecture against different mRNA lengths (21/31/41 bps) on the ground truth dataset and tested them on the independent test set. The input of different mRNA lengths does not affect the performance of the devised model. a Learning curves for the model with \(l=10, 15, 20\) (corresponding to mRNA sequence length of 21 bps, 31 bps, and 41 bps). b The ROC comparison of models with different l on the 30 split test sets from the wild-type CLASH data. c The ROC curve and performance comparison of models with different l on the independent test set