Metabolite Disease Correlation Computation

metabolite-disease-correlation-computation

Summary

Compute correlation coefficients between individual metabolite signal intensities and disease class labels across all samples, then construct and visualize a correlation matrix as a heatmap to reveal disease-associated metabolic signatures. This skill transforms raw model outputs into interpretable disease-metabolite associations.

When to use

After training a DeepMSProfiler deep learning model and generating per-sample predictions and metabolite signal intensities, use this skill when you need to: (1) identify which metabolites are most strongly associated with each disease class, (2) generate publication-ready visualizations of metabolite-disease relationships, or (3) validate that learned features correspond to biologically meaningful metabolic differences between disease groups.

When NOT to use

• Input metabolite data is already aggregated by disease group (correlation computation requires per-sample intensities and labels, not group means).
• Disease labels are continuous (e.g., disease severity score) rather than categorical; consider regression or rank correlation instead.
• Metabolite intensities contain missing values or extreme outliers that have not been preprocessed; imputation or normalization should precede correlation.

Inputs

• Per-sample model predictions (class labels for each sample)
• Per-sample metabolite signal intensities (~341 metabolites × N samples)
• Disease class labels for all samples (categorical: e.g., 'health', 'nodule', 'cancer')

Outputs

• Correlation matrix (metabolites × diseases, numeric values in [-1, 1])
• Heatmap visualization (high-resolution image file)
• Feature results array (.npy format) containing heatmap for ensemble models

How to apply

Load the per-sample model outputs (sample predictions and metabolite signal intensities) from the trained DeepMSProfiler model. Compute correlation coefficients (Pearson or Spearman) between each of the ~341 metabolite signals and disease class labels (encoded numerically) across all samples. Construct a 2D correlation matrix with metabolites as rows and disease classes as columns. Generate a heatmap visualization using Python libraries (matplotlib/seaborn) with color intensity encoding correlation magnitude and direction (typically diverging colormap: negative = blue, zero = white, positive = red). Export the heatmap as a high-resolution image file (PNG/SVG).

Related tools

• Python (matplotlib, seaborn) (Generate heatmap visualization from correlation matrix)
• DeepMSProfiler (Source tool that trains model and outputs per-sample metabolite intensities and predictions used as input to this correlation workflow) — https://github.com/yjdeng9/DeepMSProfiler
• SciPy / NumPy (Compute Pearson or Spearman correlation coefficients between metabolite signals and disease labels)

Examples

python showFeature.py

Evaluation signals

• Correlation matrix shape matches expectations: [~341 metabolites × number of disease classes], with all values in [-1, 1].
• Heatmap visual inspection: metabolites with known disease associations should show strong correlations (|r| > 0.5) in expected disease columns; background should appear neutral (near-white for correlation ≈ 0).
• No NaN or inf values in correlation matrix; all cells contain valid numeric correlation coefficients.
• Heatmap color scale is correctly oriented (red = positive correlation, blue = negative correlation, white ≈ 0) and labeled with metabolite and disease identifiers.
• Exported image file has sufficient resolution (≥300 DPI) and is readable for publication.

Limitations

• Correlation magnitude does not imply causation or biological relevance; statistically significant correlations may reflect technical artifacts or confounding.
• Pearson correlation assumes linear relationships; non-linear metabolite-disease associations will be underestimated.
• Disease class imbalance in the training set can bias correlations toward overrepresented classes.
• Raw metabolite signal intensities may contain batch effects, normalization biases, or instrument drift not accounted for in correlation; preprocessing quality affects result interpretation.

Evidence

• [other] Compute correlation coefficients (e.g., Pearson or Spearman) between each metabolite signal and disease class labels for all samples.: "Compute correlation coefficients (e.g., Pearson or Spearman) between each metabolite signal and disease class labels for all samples."
• [other] Construct a correlation matrix with metabolites as rows and diseases as columns.: "Construct a correlation matrix with metabolites as rows and diseases as columns."
• [other] Generate heatmap visualization using a Python plotting library (e.g., matplotlib, seaborn), with rows representing metabolites, columns representing diseases, and cell colors encoding correlation magnitude and direction.: "Generate heatmap visualization using a Python plotting library (e.g., matplotlib, seaborn), with rows representing metabolites, columns representing diseases, and cell colors encoding correlation"
• [readme] It takes raw metabolomics data from different disease groups as input and provides three main outputs: 1. Sample disease type labels. 2. Heatmaps depicting the correlation of different metabolite signals with diseases.: "It takes raw metabolomics data from different disease groups as input and provides three main outputs: 1. Sample disease type labels. 2. Heatmaps depicting the correlation of different metabolite"
• [readme] After run_feature ,the heatmaps were saved in ../jobs/jobs007/feature_results/ensemble_RISE.npy: "the heatmaps were saved in ../jobs/jobs007/feature_results/ensemble_RISE.npy"