Hackathon Challenges – Single Cell Multiome Data Analysis

Outlined below are some general ideas around the theme of analyzing single-cell multimodal data that can be tackled during the hackathon. Please feel free to generate your own ideas as well!

Benchmarking of current integration pipelines

  • Compare different methods (e.g., Seurat v5, Harmony, LIGER, MOFA+) on integration of unpaired multiome datasets.

  • Design integration metrics (biological conservation vs modality mixing)

Visualization of scMultiomics

Best Methods/Practices for Visualizing Different Modalities

  • Visualization techniques for unpaired data (e.g., using MultiVI to align and merge latent spaces)
    • Coverage track (ATAC) + violin (GEX) is used for paired data, are there visualizations for unpaired?
  • Dimensionality reduction techniques for visualizing high-dimensional data (e.g., UMAP, t-SNE)

Challenges in Visualizing Unpaired Data

  • Creating visual representations that highlight correlations between different modalities

Inferring True Gene Expression from Chromatin Accessibility

Data Simulation + Perturbation Testing

  • Simulate unpaired multiome data with known ground truth (e.g., synthetic paired ATAC/GEX splits) to test integration robustness.
  • Add noise or dropout in one modality to explore denoising/stability strategies.

Predictive Modeling

  • Using machine learning models to predict true gene expression from chromatin accessibility data (BABEL, MAESTRO, Bridge integration)
  • Developing interpretable models that prioritize transcription factors involved in gene regulation

Enhancer-Promoter Interactions

  • Benchmarking methods to infer gene regulatory networks from single-cell multiome data (e.g., LINGER)

Integration of Unpaired Single Cell Multiome Data

  • Best practices for dimensionality reduction while preserving biology to integrate unpaired single cell multiome data (e.g., CCA, WNN, UMAP, totalVI)
  • Benchmarking existing integration methods to evaluate their performance (e.g. GLUE, LIGER, MinNet)