Skip to content

Latest commit

 

History

History
104 lines (74 loc) · 10.9 KB

on_boarding.md

File metadata and controls

104 lines (74 loc) · 10.9 KB
graph LR
    ConfGF_Runner["ConfGF Runner"]
    Distance_Score_Model["Distance Score Model"]
    Molecular_Data_Processing["Molecular Data Processing"]
    Conformation_Generation_Evaluation["Conformation Generation & Evaluation"]
    General_Utilities["General Utilities"]
    ConfGF_Runner -- "orchestrates" --> Distance_Score_Model
    ConfGF_Runner -- "manages" --> Molecular_Data_Processing
    ConfGF_Runner -- "utilizes" --> Conformation_Generation_Evaluation
    ConfGF_Runner -- "uses" --> General_Utilities
    Distance_Score_Model -- "processes data from" --> Molecular_Data_Processing
    Distance_Score_Model -- "leverages" --> General_Utilities
    Molecular_Data_Processing -- "prepares data for" --> Distance_Score_Model
    Molecular_Data_Processing -- "uses" --> General_Utilities
    Conformation_Generation_Evaluation -- "is used by" --> ConfGF_Runner
    Conformation_Generation_Evaluation -- "relies on" --> General_Utilities
    click ConfGF_Runner href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main//ConfGF/ConfGF Runner.md" "Details"
    click Distance_Score_Model href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main//ConfGF/Distance Score Model.md" "Details"
    click Molecular_Data_Processing href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main//ConfGF/Molecular Data Processing.md" "Details"
    click Conformation_Generation_Evaluation href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main//ConfGF/Conformation Generation & Evaluation.md" "Details"
    click General_Utilities href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main//ConfGF/General Utilities.md" "Details"
Loading

CodeBoardingDemoContact

Component Details

The ConfGF project implements a framework for generating molecular conformations using a score-based generative model. The main flow involves training a Distance Score Model to learn the distribution of inter-atomic distances, which is then used by the ConfGF Runner to generate 3D molecular structures through Langevin Dynamics. Molecular data is prepared and transformed by the Molecular Data Processing component, and the quality of generated conformations is assessed by the Conformation Generation & Evaluation module. Various General Utilities support these core functionalities.

ConfGF Runner

The central orchestration component responsible for managing the entire ConfGF pipeline, including training the distance score model, evaluating its performance, and generating molecular conformations using Langevin Dynamics. It coordinates interactions between data handling, model execution, and evaluation modules.

Related Classes/Methods:

Distance Score Model

Implements the core graph neural network model, DistanceScoreMatch, which learns to predict scores related to inter-atomic distances. This model is central to the ConfGF framework, utilizing graph convolutions and transformations to process molecular graphs and output distance-based scores.

Related Classes/Methods:

Molecular Data Processing

Handles the entire lifecycle of molecular data, from loading raw SMILES strings or RDKit molecules to transforming them into graph representations suitable for the neural network. This includes managing datasets, adding higher-order edges, and incorporating chemical properties.

Related Classes/Methods:

Conformation Generation & Evaluation

Provides functionalities for generating 3D molecular conformations from distance matrices using distance geometry and for evaluating the quality of these generated conformations. It includes metrics like RMSD and MMD to assess structural accuracy.

Related Classes/Methods:

General Utilities

A collection of foundational helper functions that support various operations across the ConfGF project. This includes general PyTorch utilities for learning rate scheduling and tensor manipulation, as well as common readout functions for graph neural networks.

Related Classes/Methods:

FAQ