Skip to content

on_boarding.md

Latest commit

 

History

History
89 lines (64 loc) · 10.9 KB

on_boarding.md

File metadata and controls

89 lines (64 loc) · 10.9 KB
graph LR
    Core_Initialization_Model_Management["Core Initialization & Model Management"]
    Molecular_Feature_Engineering["Molecular Feature Engineering"]
    Molecular_Attribution_Methods["Molecular Attribution Methods"]
    SELFIES_Attribution_Orchestration["SELFIES Attribution Orchestration"]
    Core_Initialization_Model_Management -- "Provides initialized model and embedder" --> SELFIES_Attribution_Orchestration
    Molecular_Feature_Engineering -- "Generates features for model training/embedding" --> Core_Initialization_Model_Management
    Molecular_Feature_Engineering -- "Provides molecular features for attribution" --> Molecular_Attribution_Methods
    Core_Initialization_Model_Management -- "Provides trained model for predictions" --> Molecular_Attribution_Methods
    Molecular_Attribution_Methods -- "Generates molecular mutations and preliminary scores" --> SELFIES_Attribution_Orchestration
    SELFIES_Attribution_Orchestration -- "Requests feature encoding for predictions" --> Molecular_Feature_Engineering
    click Core_Initialization_Model_Management href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/mlr-xai-selfies/Core Initialization & Model Management.md" "Details"
    click Molecular_Feature_Engineering href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/mlr-xai-selfies/Molecular Feature Engineering.md" "Details"
    click Molecular_Attribution_Methods href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/mlr-xai-selfies/Molecular Attribution Methods.md" "Details"
    click SELFIES_Attribution_Orchestration href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/mlr-xai-selfies/SELFIES Attribution Orchestration.md" "Details"
Loading

CodeBoardingDemoContact

Component Details

This graph represents the architecture of the mlr-xai-selfies project, which focuses on explainable AI for molecular machine learning, particularly using SELFIES for attribution. The main flow involves initializing models and data, generating molecular features, applying various attribution methods to understand model predictions, and orchestrating the SELFIES-based attribution workflow to provide atom-level insights into molecular properties.

Core Initialization & Model Management

This component is responsible for setting up the foundational elements of the system, including reading input datasets, initializing or loading machine learning models, and optimizing their hyperparameters. It ensures that the necessary data and predictive models are prepared and optimally configured for subsequent use.

Related Classes/Methods:

Molecular Feature Engineering

This component handles the transformation of molecular structures (SMILES) into numerical features suitable for machine learning models. It provides utilities for generating various types of molecular fingerprints, which are crucial for model input.

Related Classes/Methods:

Molecular Attribution Methods

This component encompasses various techniques for attributing model predictions back to molecular substructures or atoms. It includes methods for generating mutations based on SMILES characters, individual atoms, or molecular paths, and calculating attribution scores using these mutations or SHAP.

Related Classes/Methods:

SELFIES Attribution Orchestration

This is the central control component that orchestrates the entire SELFIES-based attribution workflow. It manages the generation of SELFIES mutations, obtains predictions for original and mutated molecules, translates SELFIES-level scores to SMILES atom-level attributions, and handles canonical ordering of results, integrating various sub-processes to deliver the final attribution dataframe.

Related Classes/Methods:

FAQ