Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Graph Convolutional Networks
Extension of convolutional neural networks to graph data that applies convolution operations on molecular structures to extract hierarchical features of atoms and their bonds.
Molecular Property Prediction
Supervised learning task using GNNs to predict physicochemical, biological, or pharmacological properties of molecules from their graph structure.
Node Feature Representation
Vector encoding of atomic features including atom type, valence, electronegativity, hybridization, and other quantum properties in molecular GNNs.
Edge Feature Encoding
Vector representation of chemical bond features such as bond type (single, double, triple), aromaticity, interatomic distance, and bond angle.
Molecular Fingerprints
Binary or continuous vector representations of molecular structures generated by GNNs, capturing relevant chemical substructures for molecular similarity and similarity search.
ADMET Prediction
Application of GNNs to predict Absorption, Distribution, Metabolism, Excretion, and Toxicity properties of drug candidate compounds in the discovery process.
Quantum Chemical Properties
Properties derived from quantum mechanics calculations such as HOMO-LUMO energy, electron density, and partial atomic charges, integrated as features in molecular GNNs.
Molecular Embedding
Low-dimensional vector representation of a molecule generated by GNNs, capturing essential structural and functional features for downstream tasks.
Virtual Screening
Computational process using GNNs to rapidly evaluate millions of chemical compounds to identify potential candidates for a given therapeutic target.
Drug-Target Interaction
Task of predicting binding affinity between a compound and a protein target using GNNs to simultaneously model the molecular structure of both drug and protein.
Molecular Dynamics
Simulation of atomic and molecular movements over time, with data that can be used to train GNNs to capture flexible molecular conformations.
Chemical Space Exploration
Use of GNNs to navigate and generate new molecules in multidimensional chemical space, simultaneously optimizing multiple desired properties.
Lead Optimization
Drug discovery phase where GNNs help chemically modify promising compounds to improve their pharmacological properties while reducing toxicity.
Molecular Similarity Metrics
Quantitative measures of structural similarity between molecules based on embeddings generated by GNNs, essential for clustering and searching for analog compounds.
Pharmacophore Modeling
Identification of structural features essential to biological activity using GNNs to automatically extract pharmacophores from activity data.
Multi-task Learning in Drug Discovery
Learning approach where a single GNN simultaneously predicts multiple molecular properties, sharing representations to improve generalization and efficiency.