MTP-RBP aims to determine whether an RNA can have a binding site with a certain protein. Similar to BERT, it learns the Integrated sequence-structure-function characteristics of RNA simultaneously by pre-training.
python 3.8.8
NumPy v1.20.1
tensorflow v2.7.0
keras v2.7.0
sklearn v0.24.1
./datasets: The training and testing dataset with sequence and label indicating it is binding sites or not
./model: The MTP-RBP model code, which consists of a total of 6 parts :
- Part1：Pre-training mask data processing
- Part2：Model Building Functions
- Part3：Pre-training Step 1: Train MLM and BFP pre-training tasks
- Part4：Pre-training Step 2: Train MLM, BFP and SSC pre-training tasks
- Part5：Fine tuning
- Part6：Evaluation of fine-tuning results
./structBEP: BEP Representation of Structures Used to Guide Secondary Structure Construction Tasks
For RNA secondary structure prediction, we use RNAfold to implement the prediction of RNA secondary structure. The predicted secondary structure is represented in a dot-bracket format, where each symbol corresponds to a base site in the sequence. Following the principle of base complementary pairing yields a secondary structure pairing tree diagram.
You can use RNAfold to make predictions about RNA secondary structure in the following ways
conda install -c bioconda viennarna
RNAfold --noPS < input.fa > output.faYou can choose the same tool or other RNA secondary structure prediction tools.
Contact: Lin Gan ([email protected])