MultiModRLBP: A Deep Learning Approach for RNA-Small Molecule Ligand Binding Site Prediction using Multi-modal features

This repository is an implementation of RNA-Small Molecule Ligand Binding Site Prediction by deep learning approach.

We implement four major components:

• data: Dataset
• Model: Complete model and training code
• RGCN: RGCN Module
• RNABert: RNABert Module

See README in each folder for details on how to use each component.

Requirements

• Python >= 3.6
• Pytorch
• dgl
• OpenBabel
• BioPython
• tqdm
• rnaglib
• networkx >= 2.1

You can automatically install all the dependencies with Anaconda using the following command:

conda env create -f environment.yml

Usage

Loading a trained model

• You can use the model used for the paper, or load a trained model you trained yourself (see next section)

Making predictions for every graph in a folder.

Create a script in the root of the repository with the following code:

from RGCN.data_loading import graphloader
from RGCN.benchmark import evaluate
from RGCN.kernels import node_sim
from RnaBert.MLM_SFP import  get_config, TRAIN, BertModel, BertForMaskedLM\

config = get_config(file_path = "RNA_bert_config.json")
config.hidden_size = config.num_attention_heads * config.multiple    
train = TRAIN(config,device)
RnaBert = BertModel(config)
RnaBert = BertForMaskedLM(config, RnaBert)
RnaBert = train.model_device(RnaBert)
RnaBert.load_state_dict(torch.load("bert_mul_2.pth"))
embedder_model = model.RGATEmbedder(infeatures_dim=supervised_train_dataset.input_dim+2 ,
                                        dims=[64, 64])
classifier_model = model.RGATClassifier(rgat_embedder=embedder_model,rbert_embedder=RnaBert, 
                                            conv_output=False,
                                            return_loss=False,
                                            classif_dims=[supervised_train_dataset.output_dim])
model_dict = torch.load(args.load_path)

classifier_model.load_state_dict(model_dict['model_state_dict'])

Predict

auc, mcc, precision,  recall = evaluate.get_performance(node_target=node_target, node_features=node_features, model=classifier_model,test_loader=test_loader)

Training your own model

For a full list of command-line options:

$ python MultiModRLBP_train.py -h