SSCLMD

Accept journal IEEE Journal of Biomedical and Health Informatics ## Paper [Paper link]

1. Overview

The code for paper Self-supervised contrastive learning on attribute and topology graphs for predicting relationships among lncRNAs, miRNAs and diseases". The repository is organized as follows:

• data/ contains the dataset 1 and dataset 2 used in the paper, with dataset 1 as an example;
  • lnc(mi)_dis_association_new2.txt and lnc_mi_interaction_new2.txt contain known lncRNA(miRNA)-disease associations and lncRNA-miRNA interactions, respectively;
  • LDA/MDA/LMI.edgelist contain known LDA, MDA, and LMI pairs, respectively; no_LDA/MDA/LMI.edgelist contain unknown LDA, MDA, LMI pairs;
  • lncRNA/miRNA_sequences2.xlsx contain lncRNA and miRNA sequences, lncRNA sequences are from NCBI, miRNA sequences are from miRBase;
  • disease_name.xlsx contains disease names and their DOID numbers;
  • dis_sem_sim.txt contains disease semantic similarity data:
• code/
  • data_preparation.py is used to calculate lncRNA/miRNA k-mer features and construct knn graph (attribute graph) of lncRNA/miRNA/disease.
  • calculating_similarity.py is used to calclulate lncRNA/miRNA/disease GIPK similarities and obtain the intra-edges in the topology graph;
  • parms_setting.pycontains hyperparmeters;
  • utils.py contains preprocessing function of the data;
  • data_preprocess.py contains the preprocess of data;
  • layer.py contains SSCLMD's model layer;
  • train.py contains training and testing code;
  • main.py runs SSCLMD;

2. Dependencies

• numpy == 1.21.1
• torch == 2.0.0+cu118
• sklearn == 0.24.1
• torch-geometric == 2.3.0

3. Quick Start

Here we provide a example to predict the lncRNA-disease association scores on dataset 1:

1. Download and upzip our data and code files
2. Run data_preparation.py and calculating_similarity.py to obtain lncRNA/miRNA/disease attribute graph and intra_edge of topology graph
3. Run main.py (in file-- dataset1/LDA.edgelist, neg_sample-- dataset1/non_LDA.edgelist, task_type--LDAl)

4. Reminder

It is recommended that you save the training and test sets for each fold and then calculate the lncRNA/miRNA/disease functional similarity. Then continue with subsequent calculations, which will speed up the calculation.

5. Contacts

If you have any questions, please email Nan Sheng (shengnan@jlu.edu.cn)