graph LR
AESTETIK_API_Orchestrator["AESTETIK API & Orchestrator"]
Data_Preparation_Loading["Data Preparation & Loading"]
Core_Autoencoder_Model["Core Autoencoder Model"]
PyTorch_Lightning_Training_Module["PyTorch Lightning Training Module"]
Loss_Function_Metrics["Loss Function & Metrics"]
Post_processing_Analysis["Post-processing & Analysis"]
Visualization_Utilities["Visualization Utilities"]
AESTETIK_API_Orchestrator -- "triggers" --> Data_Preparation_Loading
AESTETIK_API_Orchestrator -- "configures" --> PyTorch_Lightning_Training_Module
Data_Preparation_Loading -- "provides data to" --> PyTorch_Lightning_Training_Module
PyTorch_Lightning_Training_Module -- "interacts with" --> Core_Autoencoder_Model
PyTorch_Lightning_Training_Module -- "sends outputs and targets to" --> Loss_Function_Metrics
Loss_Function_Metrics -- "returns computed loss to" --> PyTorch_Lightning_Training_Module
Core_Autoencoder_Model -- "outputs latent space and reconstructed data to" --> PyTorch_Lightning_Training_Module
PyTorch_Lightning_Training_Module -- "outputs trained model and latent space to" --> AESTETIK_API_Orchestrator
AESTETIK_API_Orchestrator -- "sends latent space to" --> Post_processing_Analysis
Post_processing_Analysis -- "sends results to" --> AESTETIK_API_Orchestrator
AESTETIK_API_Orchestrator -- "sends analysis results to" --> Visualization_Utilities
Visualization_Utilities -- "sends generated visualizations to" --> AESTETIK_API_Orchestrator
The AESTETIK library orchestrates a comprehensive spatial transcriptomics analysis pipeline. It begins with the AESTETIK API & Orchestrator initiating Data Preparation & Loading to process raw data, including calibration and grid construction. This prepared data is then fed into the PyTorch Lightning Training Module, which encapsulates the Core Autoencoder Model (comprising encoder and decoder) and interacts with Loss Function & Metrics during training. After training, the AESTETIK API & Orchestrator retrieves the trained model and latent space, passing the latter to Post-processing & Analysis for clustering. Finally, the results are channeled to Visualization Utilities to generate insightful plots, completing the analytical workflow.
The central control unit, managing the entire AESTETIK workflow from data input to result visualization. It provides the user interface and coordinates interactions between all other components.
Related Classes/Methods:
Responsible for ingesting raw spatial transcriptomics data, performing necessary preprocessing steps (e.g., calibration, grid building), and creating efficient data loaders for model consumption.
Related Classes/Methods:
Implements the autoencoder neural network architecture, consisting of an encoder for dimensionality reduction and a decoder for data reconstruction. This is the core machine learning component.
Related Classes/Methods:
Encapsulates the Core Autoencoder Model within the PyTorch Lightning framework, handling training, validation, and testing loops, and integrating with optimization and loss calculation.
Related Classes/Methods:
Provides the mathematical functions (e.g., reconstruction loss, triplet loss) used to guide the autoencoder's learning process and evaluate its performance.
Related Classes/Methods:
Performs downstream analysis on the latent space generated by the autoencoder, primarily focusing on clustering to identify biological patterns.
Related Classes/Methods:
Generates various plots and visual representations of the AESTETIK analysis results, aiding in the interpretation of biological insights.
Related Classes/Methods: