mTopic - Multimodal topic modeling for single-cell data
=======================================================

`mTopic` is an open-source Python library with computational tools for modeling multimodal topics in single-cell data.

.. raw:: html

    <div class="mTopic-logo-container">
      <img class="logo-light"
           src="_static/mTopic_logo_light_background.png"
           alt="mTopic logo (light)">
      <img class="logo-dark"
           src="_static/mTopic_logo_dark_background.png"
           alt="mTopic logo (dark)">
    </div>

    <script>
      function swapLogo() {
        const theme = document.documentElement.getAttribute('data-theme');
        const isDark = theme === 'dark';
        const lightImg = document.querySelector('.mTopic-logo-container .logo-light');
        const darkImg  = document.querySelector('.mTopic-logo-container .logo-dark');
        if (lightImg && darkImg) {
          lightImg.style.display = isDark ? 'none'  : 'block';
          darkImg .style.display = isDark ? 'block' : 'none';
        }
      }

      // Initial swap after all scripts run
      window.addEventListener('load', swapLogo);
      // Watch for theme‐toggle changes
      new MutationObserver(swapLogo)
        .observe(document.documentElement, {
          attributes: true,
          attributeFilter: ['data-theme', 'data-bs-theme']
        });
    </script>

mTopic is a scalable Python package for topic modeling on multimodal single-cell datasets. It supports spatial and non-spatial data, enabling a joint analysis across multiple molecular layers, such as RNA, ATAC, and protein expression.

By capturing shared patterns across modalities, mTopic helps reveal latent biological structures, regulatory programs, and spatial organization within complex tissues.

Resources
---------

- GitHub (Python) (https://github.com/TabakaLab/mTopic)  
- GitHub (R companion) (https://github.com/TabakaLab/mTopicR)  
- Documentation (https://mtopic.readthedocs.io/)

Features
--------

- Supports simultaneous inference of multimodal molecular programs from any set of single-cell modalities
- Handles spatial and non-spatial single-cell data
- Implements scalable variational inference for efficient training
- Allows per-modality preprocessing, signature detection, and z-score enrichment
- Built on top of the MuData (https://muon.readthedocs.io/) data structure (`.h5mu` format)
- Visualization tools for topic distributions, dominant topic maps, and feature signatures

.. toctree::
   :caption: Documentation
   :hidden:

   installation
   notebooks/T1_P22_Mouse_Brain_preprocessing
   notebooks/T2_P22_Mouse_Brain_training
   notebooks/T3_Human_Tonsil_training
   notebooks/T4_Human_PBMC_training
   notebooks/T5_Topic_Number_Selection
   api