C-COMPASS: a user-friendly neural network tool profiles cell compartments at protein and lipid levels

Daniel T. Haas, Daniel Weindl, Pâmela A. Kakimoto, Eva-Maria Trautmann, Julia P. Schessner, Xia Mao, Mathias J. Gerl, Maximilian Gerwien, Timo D. Müller, Christian Klose, Xiping Cheng, Jan Hasenauer, Natalie KrahmerNature Methods 2025 · DOI

TL;DR

This paper introduces C-COMPASS, a user-friendly software tool that uses a neural network to predict the precise location of proteins and lipids within cellular compartments. This advancement allows for a more comprehensive understanding of cellular organization and dynamics, even when proteins or lipids reside in multiple locations, and integrates proteomic and lipidomic data for a holistic view.

What Problem Does This Paper Address?

Understanding how cells organize themselves into distinct compartments is fundamental to biology. While techniques like protein correlation profiling (PCP) and localization of organelle proteins by isotope tagging (LOPIT) have helped map protein locations, analyzing this data is complex. Many proteins and lipids don't reside in just one place (multilocalization), and existing tools struggle to quantify these distributions or integrate different types of molecular data. Furthermore, identifying lipid locations is particularly challenging due to the lack of specific lipid markers for organelles. This paper addresses the need for a more accessible, quantitative, and multi-omic approach to mapping cellular compartments.

Methods

The core of C-COMPASS is a neural network-based regression model. It takes protein abundance data across different cellular fractions (obtained from techniques like density gradient centrifugation) and a list of known marker proteins for specific compartments as input. The network is trained to predict the "class contribution" (CC) for each protein in each compartment, representing its quantitative distribution.

Key innovations include:

  • Handling Multilocalization: Unlike traditional methods that often assign a single location, C-COMPASS quantifies the percentage of a protein found in each compartment.
  • Synthetic Data Generation: To address imbalances in marker protein availability and create training data for mixed localizations, the tool generates artificial protein profiles.
  • Lipid Profiling Integration: By training the network with protein marker profiles, C-COMPASS can also predict lipid localization, overcoming the scarcity of organelle-specific lipid markers.
  • User-Friendly Interface: C-COMPASS offers a graphical user interface, making it accessible to researchers without extensive computational expertise.

Datasets

The paper utilizes several datasets for training and evaluation:

  • Human White Adipocyte Data: A published dataset from a PCP experiment was used to benchmark C-COMPASS against existing Support Vector Machine (SVM) methods.
  • HyperLOPIT Dataset: Data from a study using differential centrifugation and tandem mass tag labeling in THP-1 cells was used to assess C-COMPASS's applicability across different fractionation workflows.
  • Humanized Mouse Liver Data: Proteomic and lipidomic data from mice with humanized livers under different metabolic conditions (chow diet, high-fat/high-fructose diet, fasting) were generated and analyzed.
  • Simulated Datasets: Synthetic datasets were created to systematically evaluate C-COMPASS's performance under controlled conditions, varying fraction numbers, compartment numbers, and localization patterns.

Key Insights and Ideas

  • Beyond Single Localization: The ability to quantitatively predict multilocalization is crucial for understanding cellular complexity. Many proteins and lipids are not confined to a single organelle, and C-COMPASS provides a way to measure this distribution.
  • Lipid Mapping Breakthrough: Integrating lipidomics with proteomics opens up a new frontier in organelle mapping. This allows researchers to understand how the lipid composition of organelles changes and how it relates to protein function.
  • Dynamic Cellular Architecture: By analyzing cells under different metabolic conditions, C-COMPASS can reveal how organelle composition and protein/lipid localization change in response to physiological stimuli, offering insights into cellular remodeling.
  • Democratizing Spatial Proteomics: The user-friendly interface and open-source nature of C-COMPASS lower the barrier to entry for complex spatial proteomics and lipidomics analyses.

Key Results

  • Improved Prediction Accuracy: C-COMPASS outperformed traditional SVM-based methods in predicting protein localization, particularly for underrepresented organelles, achieving an F1 score of 0.84 compared to SVM's 0.80 on a benchmark dataset.
  • Robustness Across Workflows: The tool demonstrated high prediction reliability (average F1 score of 0.96) across different organelle separation methods, including PCP and HyperLOPIT.
  • Identification of Novel Relocalizations: In a humanized mouse liver model, C-COMPASS identified previously uncharacterized fasting-induced protein relocalizations, such as those involving accessory subunits of the histone deacetylase SIN3-HDAC complex.
  • Quantitative Lipid Localization: The study generated the first systematic organelle map at the lipid level, identifying organelle-specific lipid species and variations in membrane lipid composition. For example, it showed distinct patterns of fatty acid chain length and unsaturation in lipids across different organelles.
  • Performance with Fraction Resolution: The accuracy of C-COMPASS predictions improved with an increasing number of fractions, with optimal performance achieved when the number of fractions was at least twice the number of compartments.

Limitations and Future Work

  • Resolution Dependence: C-COMPASS's accuracy, like other fractionation-based methods, depends on the resolution of the experimental data. Compartments with overlapping fractionation profiles (e.g., Golgi and LDs) may present challenges.
  • Data Quality: The quality of the input fractionation data is critical for accurate predictions.
  • Imputation Strategy: The current version uses zero imputation for missing values, which is appropriate for fractionation data but could be expanded with alternative strategies.
  • Future Directions: The authors suggest extending C-COMPASS to metabolic tracing approaches for lipid flux analyses and further developing its capabilities for dynamic processes and multi-omic integration.

Summary updated

Abstract

Systematic proteomic organelle profiling methods including protein correlation profiling and LOPIT have advanced our understanding of cellular compartmentalization. To manage the complexity of organelle profiling data, we introduce C-COMPASS, a user-friendly open-source software that employs a neural network-based regression model to predict the spatial cellular distribution of proteins. C-COMPASS handles complex multilocalization patterns and integrates protein abundance to model organelle composition changes across conditions. We apply C-COMPASS to mice with humanized livers to elucidate organelle remodeling during metabolic perturbations. Additionally, by training neural networks with co-generated marker protein profiles, C-COMPASS extends spatial profiling to lipids, overcoming the lack of organelle-specific lipid markers, allowing for determination of localization and tracking of lipid species across different compartments. This provides integrated snapshots of organelle lipid and protein compositions. Overall, C-COMPASS offers an accessible tool for multiomic studies of organelle dynamics without needing advanced computational skills, empowering researchers to explore new questions in lipidomics, proteomics and organelle biology.
BibTeX
@inproceedings{haas2025c,
  author = {Daniel T. Haas and Daniel Weindl and Pâmela A. Kakimoto and Eva-Maria Trautmann and Julia P. Schessner and Xia Mao and Mathias J. Gerl and Maximilian Gerwien and Timo D. Müller and Christian Klose and Xiping Cheng and Jan Hasenauer and Natalie Krahmer},
  title = {{C-COMPASS: a user-friendly neural network tool profiles cell compartments at protein and lipid levels}},
  booktitle = {Nature Methods},
  year = {2025},
  doi = {10.1038/s41592-025-02880-3},
}
Figure