LM-DTI

LM-DTI predicts drug-target interactions by integrating drugs, protein targets, long non-coding RNAs (lncRNAs), and microRNAs (miRNAs) within a heterogeneous information network to support drug function discovery and repositioning.

Key Features:

• lncRNA and miRNA integration: Integrates long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) into the DTI prediction framework.
• Heterogeneous information network: Constructs a heterogeneous information network comprising eight distinct networks with four node types: drugs, targets, lncRNAs, and miRNAs.
• Graph embedding (node2vec): Applies node2vec graph embedding to generate feature vectors for drug and target nodes.
• Network path scoring (DASPfind): Calculates path score vectors for each candidate drug-target pair using DASPfind to assess connectivity and interaction pathways.
• Feature integration and classification (XGBoost): Combines node2vec feature vectors and DASPfind path score vectors and uses an XGBoost classifier to predict DTIs.
• Performance evaluation: Reported performance by 10-fold cross-validation with an Area Under Precision-Recall Curve (AUPR) of 0.96.
• Validation: Validation performed via manual literature searches and database verifications.
• Computational characteristics: Reported computational scalability and efficiency for handling large heterogeneous networks.

Scientific Applications:

• Identifying novel drug-target interactions: Enables discovery of previously unrecognized DTIs.
• Drug function discovery: Supports characterization of drug mechanisms through predicted target interactions.
• Drug repositioning: Facilitates identification of new indications for existing drugs based on predicted DTIs.

Methodology:

Constructs a heterogeneous information network (eight networks, four node types), generates node embeddings for drugs and targets using node2vec, computes path score vectors per drug-target pair with DASPfind, combines embeddings and path scores and inputs them to an XGBoost classifier, and evaluates predictions by 10-fold cross-validation reporting AUPR.