Examples

This section contains examples to show how to prepare the data for UNAGI running, analyse the UNAGI outcomes, perform perturbations and visualize the reuslts using the example IPF dataset.

Running UNAGI on the example dataset

• Dataset preparation
• Configure the model architecture and training parameters
  • Part 1: Setup and load the datasets
  • Part 2: Configure the model architecture of UNAGI and training hyper-parameters
  • Part 3: Perform in-silico perturbations and downstream analysis
• Visualize the cell type composition acorss stages
• Visualize the cell embeddings colouring by cell types
• Plot the dotplots for cell type markers at individual stages
• Plot dynamic markers for a specific track
• Plot the dendrogram of each stage data
• Visualize hierarchical markers using heatmaps
• Visualize the pathway perturabtion results
• Visualize the drug/compounds perturabtion results

UNAGI ablations

• Ablations and embedding quality evaluation
  • UNAGI w.o GAN and GCN
  • UNAGI w.o GCN
  • UNAGI w.o. GAN
  • Evaluate embedding quality

Using UNAGI to predict post-treatment gene expressions

• Train UNAGI model based on the PCLS data
• Perform in-silico perturbation and visualize the predicted post-treatment gene expression
• Perform the in-silico perturbation and calcualte pearson correlation of predicted gene expression changes

Using UNAGI with a customized pathway or drug database for in-silico perturbation

• In-silico perturbation for customized pathway databases
  • Built-in pathway databases
  • Customize pathway database
  • Perform in-silico pathway perturbation on the customized pathway database
• In-silico perturbation for customized drug(compound) databases
  • Customize drug (compound) databases
  • Perform in-silico drug perturbation on the customized drug database

UNAGI walkthrough

• UNAGI walkthrough
  • Install UNAGI from pypi
  • Use the example dataset
  • Load UNAGI
  • load the dataset
  • Configure the model architecture of UNAGI and training hyper-parameters
  • Load the post-analysis data
  • Plot the composition of cells in individual stages
  • Plot the UMAPS of cell embeddings for individual stages
  • Plot dotplots of cell type markers
  • Plot increasing gene markers and decreasing gene markers from a specific track
  • Plot dendrogram of hierarchical clusterings
  • Plot hierarchical static markers heatmap
  • Perturbation using customized pathway database
  • Customized drug perturbation