MANTIS: Analytics toolkit for spatial metabolomics with matching spatial transcriptomics data

MANTIS is a novel statistical framework that unifies spatial metabolomics and transcriptomics data to rigorously identify gene-metabolite relationships and spatial patterns while accounting for spatial dependence and confounding factors through autocorrelation-preserving permutation strategies.

The Gist

Imagine you are a detective trying to solve a mystery inside a bustling, microscopic city: a piece of living tissue.

In this city, there are two main types of residents you want to understand:

1. The "Blueprints" (Genes/Transcriptomics): These are the instructions telling the cells what to do.
2. The "Supplies" (Metabolites): These are the actual chemicals, fuels, and building blocks the cells are using right now.

For a long time, scientists could look at the blueprints or the supplies, but rarely both at the same time in the exact same spot. Recently, technology advanced enough to let us see both maps simultaneously. But here's the problem: We didn't have a good way to read them together.

Existing tools were like using a magnifying glass to look at a single street corner. They could tell you what was on that corner, but they couldn't tell you if the pattern of supplies was just because of the neighborhood, or if it was actually linked to the blueprints. They often missed the big picture or got confused by the layout of the city.

Enter MANTIS: The Master Detective's Toolkit

The paper introduces MANTIS (Metabolomics And Transcriptomics In Space). Think of MANTIS not just as a tool, but as a super-smart detective that has three special superpowers to solve the tissue mystery.

1. The "Neighborhood Watch" (Finding Regional Patterns)

Imagine you walk through a city and notice that everyone in the "Financial District" is wearing suits, while everyone in the "Art District" is wearing bright colors.

• The Old Way: Tools would just say, "Hey, suits are common here!" but they might get confused if the whole city was just wearing suits.
• MANTIS: It asks, "Is this pattern special to this neighborhood, or is it just random?" It uses a clever trick (called a "permutation strategy") to shuffle the data around like a deck of cards while keeping the neighborhood's shape intact. If the pattern still stands out after shuffling, MANTIS knows, "Aha! This chemical is specifically for this neighborhood!"

2. The "Cell-Type Matchmaker" (Finding Cell Associations)

Sometimes, a chemical isn't about the neighborhood; it's about the type of person living there. Maybe only the "Bakers" (a specific cell type) use a lot of flour, regardless of which street they are on.

• The Old Way: Tools might just count how much flour is in a house and say, "High flour!" without realizing it's because a Baker lives there.
• MANTIS: It acts like a matchmaker. It looks at the "Bakers" and the "Flour" and asks, "Do they hang out together more than you'd expect by chance?" It uses a "Spatial Cross-Correlation" score, which is like measuring how often two friends are seen walking down the street together, rather than just being in the same city.

3. The "Truth Filter" (Removing the Confusion)

This is MANTIS's most powerful trick. Sometimes, a Gene and a Chemical look like they are best friends because they both live in the "Downtown" district. But they might not actually know each other; they just live in the same zip code!

• The Problem: If you just look at the data, you might think Gene A and Chemical B are a perfect couple.
• MANTIS: It uses a "Spatial Partial Correlation" filter. It's like saying, "Okay, let's pretend they don't live in Downtown. If they still hang out together, then they are a real match."
  • It strips away the "Neighborhood" effect.
  • It strips away the "Cell Type" effect.
  • If the connection remains, MANTIS says, "This is a genuine biological relationship!"

Why Does This Matter? (The Real-World Impact)

The authors tested MANTIS on real data, like a map of a mouse brain and a human lung tumor.

• In the Brain: They found that dopamine (a happy chemical) and certain proteins were negatively linked. Why? Because dopamine lives in the "gray matter" (neurons) and the proteins live in the "white matter" (myelin). MANTIS realized they weren't fighting each other; they were just living in different parts of the city. Once MANTIS accounted for the "city zones," it found other chemicals that were actually interacting in interesting ways.
• In Cancer: They found that immune cells (T-cells) were closely linked to specific fats (ceramides) in the tumor. This suggests the immune system is actively remodeling the tumor's environment, a clue that could help doctors design better treatments.

The Bottom Line

Before MANTIS, looking at genes and chemicals together in space was like trying to listen to a crowded party by only hearing the loudest voices. You missed the subtle conversations.

MANTIS is the noise-canceling headphones. It filters out the background noise of "where things are" and "what kind of cells they are," so scientists can finally hear the true, specific conversations between genes and chemicals. This helps us understand how our bodies work, how diseases like cancer or Parkinson's develop, and how to fix them.

Technical Summary

1. Problem Statement

The integration of Spatial Metabolomics (SM) and Spatial Transcriptomics (ST) offers a powerful multi-omics view of tissue function, allowing researchers to map metabolic states to specific cell types and spatial domains. However, current computational tools for analyzing matched SM+ST data suffer from significant limitations:

• Lack of Spatial Awareness: Existing methods (e.g., SpaMTP, SpatialMETA) often rely on non-spatial correlation metrics (like Pearson correlation) to link genes and metabolites. This fails to account for the inherent spatial autocorrelation in tissue data, leading to inflated false discovery rates.
• Inability to Disentangle Confounders: Current tools struggle to distinguish whether a gene-metabolite association is driven by direct biological interaction or merely by co-localization within a specific tissue region or cell type.
• Limited Statistical Rigor: There is a lack of robust statistical frameworks that explicitly model spatial dependence and provide calibrated significance testing for spatial patterns and associations.

2. Methodology: The MANTIS Framework

MANTIS (Metabolomics And Transcriptomics In Space) is a Python-based statistical framework designed to analyze co-registered SM+ST data at single-cell or spot resolution. It assumes input data includes spatial coordinates, metabolite abundances, gene expression, and optional annotations for cell types and spatial domains.

The toolkit employs two primary categories of analysis:

A. Discovery of Metabolic Spatial Patterns

MANTIS identifies three distinct classes of metabolite patterns:

1. Region-Associated Metabolites (regional_met):
   • Uses Kullback-Leibler (KL) divergence to measure how a metabolite's abundance distribution across spatial domains deviates from a background profile.
   • Significance Testing: Uses a novel autocorrelation-preserving permutation strategy (Simulated Annealing sampler) to generate null distributions. This preserves the spatial structure of the data while randomizing values, ensuring rigorous control over spatial autocorrelation.
2. Cell Type-Associated Metabolites (celltype_met):
   • Utilizes the Spatial Cross-Correlation Index (SCI), a bivariate Moran's I statistic, to quantify the spatial covariance between a metabolite's abundance and cell type presence (or proportions).
   • Significance is assessed via the same autocorrelation-preserving permutation approach.
3. Spatially Variable Metabolites (spatvar_met):
   • Detects metabolites with spatial patterns that cannot be explained by known regions or cell types.
   • Confounder Correction: First regresses metabolite abundance against spatial domain or cell type information (using OLS or LASSO) and calculates Moran's I on the residuals. This isolates "true" spatial variability from domain-driven clustering.

B. Discovery of Gene-Metabolite Associations

MANTIS moves beyond simple correlation to identify specific molecular relationships:

• Spatial Cross-Correlation (SCI): Calculates the spatial co-variation between gene expression and metabolite abundance.
• Spatial Partial Correlation (SPC): To avoid spurious correlations caused by co-localization, MANTIS introduces SPC-CT (adjusting for cell type) and SPC-SD (adjusting for spatial domains).
  • The method fits gene and metabolite levels to the confounder (e.g., cell type), extracts residuals, and calculates the SCI on these residuals.
  • This isolates associations that are not merely artifacts of both molecules being present in the same cell type or region.

3. Key Contributions

• First Unified Framework: MANTIS is the first toolkit to unify spatial metabolomics, transcriptomics, cell type, and spatial domain information within a single statistical framework emphasizing hypothesis testing.
• Autocorrelation-Preserving Permutation: The introduction of a Simulated Annealing (SA) sampler to generate null distributions that preserve spatial autocorrelation is a major methodological advancement. This provides calibrated inference that existing tools lack.
• Confounder Disentanglement: The toolkit explicitly separates associations driven by tissue organization (regions/cell types) from direct gene-metabolite regulatory relationships using Spatial Partial Correlation.
• Spatially Aware Statistics: Replaces standard Pearson correlation with SCI and SPC, ensuring that findings are robust to the spatial covariance structures inherent in biological tissues.

4. Results and Validation

The authors validated MANTIS on three datasets: a mouse striatum (VICARI), a human lung cancer dataset, and a human Parkinson's disease striatum dataset.

• Mouse Striatum (VICARI):
  • Identified 66 significant region-associated metabolites (FDR < 0.1), a much more conservative and specific set compared to existing tools (which reported thousands of false positives due to lack of spatial correction).
  • Discovered 1,518 cell type-metabolite associations, with strong links to oligodendrocytes and glial cells.
  • Found 73 spatially variable metabolites that remained significant after regressing out domain and cell type effects, revealing unique gradients and focal hotspots.
  • Gene-Metabolite Analysis: Initial SCI analysis found 12.3% significant pairs. After applying SPC-CT and SPC-SD, only 2.4% of pairs remained significant, demonstrating that the majority of initial correlations were driven by shared spatial localization rather than direct interaction.
• Human Lung Cancer:
  • Applied to single-cell proteomics/metabolomics data. Identified links between T-cell markers (CD45RA/RO) and specific lipid metabolites, suggesting immune infiltration drives local metabolic remodeling.
  • Confirmed elevated glycine in tumor regions, consistent with known one-carbon metabolism upregulation in cancer.
• Human Parkinson's Striatum:
  • Revealed a strong negative spatial correlation between Dopamine and myelin proteins (Mbp, Plp1), driven by the segregation of gray vs. white matter.
  • After correcting for spatial domains, a specific module of gene-metabolite pairs (e.g., Cntn2 and GABA-H2O) remained significant, highlighting inverse spatial relationships between myelinated axons and GABAergic neurons.

5. Significance

MANTIS represents a critical step forward in spatial multi-omics analysis by shifting the paradigm from descriptive clustering to rigorous statistical inference.

• Reduced False Positives: By accounting for spatial autocorrelation and confounders, MANTIS prevents the misinterpretation of co-localization as biological causality.
• Biological Interpretability: The ability to disentangle "where" a metabolite is found (region/cell type) from "what" it interacts with (gene-metabolite pairs) allows for more precise mechanistic hypotheses.
• Broad Applicability: The toolkit is platform-agnostic, working across different SM technologies (MALDI-MSI, DESI) and ST platforms (Visium, single-cell), making it a versatile tool for the growing field of spatial biology.

In summary, MANTIS provides the necessary statistical rigor to unlock the full potential of joint SM+ST profiling, enabling researchers to move beyond simple mapping to the discovery of causal and regulatory spatial mechanisms.