Welcome to methBERT’s documentation
MethBERT explores a non-recurrent modeling approach for nanopore methylation detection based on the bidirectional encoder representations from transformers (BERT). Compared with the state-of-the-art model using bi-directional recurrent neural networks (RNN), BERT can provide a faster model inference solution without the limit of computation in sequential order. We proviede two types of BERTs: the basic one [Devlin et al.] and the refined one. The refined BERT is refined according to the task-specific features, including:
learnable postional embedding
self-attetion with realtive postion representation [Shaw et al.]
center postitions concatenation for the output layer
(Currently, we only trained on the R9 benchmark data. R9.4.1 and R10.3 models will be provided in the next update, when our data is ready.)
Reference
Yao-zhong Zhang et al., On the application of BERT models for nanopore methylation detection
Liu et al. DeepMod
Devlin et al., https://arxiv.org/pdf/1810.04805.pdf
Shaw et al., https://arxiv.org/pdf/1803.02155.pdf