GACNNMDA

GACNNMDA predicts potential human microbe-drug associations using graph attention networks (GANs) and convolutional neural networks (CNNs).

Key Features:

• Heterogeneous Network Construction: Constructs two distinct heterogeneous microbe-drug networks based on multiple similarity measures, known associations, and microbe-disease-drug relationships.
• Feature Matrix Integration: Integrates various attributes of microbes and drugs into comprehensive feature matrices.
• Graph Attention Network Utilization: Inputs the constructed networks and feature matrices into a two-layer graph attention network (GAN) to derive low-dimensional feature representations for microbes and drugs.
• Convolutional Neural Network Application: Combines low-dimensional representations with their respective feature matrices as inputs to a convolutional neural network (CNN) to predict association scores for microbe-drug pairs.

Scientific Applications:

• Drug discovery and development: Predicts candidate microbe-drug associations to inform identification of new drug targets and optimization of existing treatments.
• Prioritization for experimental follow-up: Ranks microbe-drug pairs by predicted association scores to guide experimental validation studies.

Methodology:

Construct two heterogeneous microbe-drug networks from multiple similarity measures, known associations, and microbe-disease-drug relationships; concatenate attributes into feature matrices for microbes and drugs; input networks and feature matrices into a two-layer graph attention network to obtain low-dimensional representations; combine those representations with feature matrices and input them to a convolutional neural network to predict microbe-drug association scores.