Source code for paper: AttentionSplice: An interpretable multi-head self-attention based hybrid deep learning model in splice site prediction
Detecting splice sites is important for current DNA annotation system and challenges the conventional methods. We propose the AttentionSplice model, which combines multi-head self-attention and hybrid deep learning construction to identify the splice sites. We extract important positions and key motifs which could be essential for splice site detection.
We conduct our experiments on the following environments:
python == 3.7
torch == 1.8.0
cuda == cu111
transformer == 4.7.0
gpu: GeForce RTX 3090
we use Human Nuclear DNA sequence data and annotations
of the corresponding sequences were acquired from HS3D.you can find this dataset in http://www.sci.unisannio.it/docenti/rampone/
we also use the caenorhabditis elegans dataset (CED), you can find this dataset in https://public.bmi.inf.ethz.ch/user/behr/splicing/C_elegans/
- You can put the data files in the datasets folder
- run p_attention in /attention/pytorch_attention
python p_attention.py- log_file will save under pytorch_attention folder