Mapping Adipocyte Interactome Networks with HALOTAG Enrichment Mass Spectrometry

This repository provides the methodology used in our study to visualize latent variables and protein interaction scores for protein structures using AlphaFold2-based Localcolabfold.

Preparation

1. Install Localcolabfold
   Install Localcolabfold in your local environment.

2. Load Target Protein Sequences
   Retrieve the target protein sequences from UniProt.

Setup

$ poetry install

Calculation (Localcolabfold)

1. Predict Protein Interaction Structures
   Calculate the structure predictions for protein interactions. (localcolabfold/colab-run-multi-seed0.sh, localcolabfold/get_score_multi.py)

2. Predict Individual Protein Structures and Extract Latent Variables
   Calculate the structure predictions for individual proteins and extract latent variables (256 dimensions) from the model. (localcolabfold/colab-run-mono-seed0.sh, localcolabfold/af2emb.py)

3. Merge Data
   Merge the latent variables of proteins, the structure prediction scores from protein interactions, and the pathway data from KEGG. (localcolabfold/merge_score_af2emb.py)

*Note: The calculations in step 1, 2 and 3 are extensive, and we provide the calculated results. (data)

Visualization

1. Dimensionality Reduction and Visualization
   Using UMAP, reduce the dimensionality of the latent variables to two dimensions for interacting proteins with each bait protein used in the Halo-MS method. Visualize these reduced variables together with the interaction scores. (Run : visualization/makefig_adsc.ipynb / Results : figures)

   *Note: The figure generated here is used in the paper.