LobeHub - Empowering your AI dreams

Purpose

Imputes missing/dropout values in scRNA-seq expression data to improve metabolic pathway analysis. This process handles sparsity common in single-cell RNA sequencing data by filling in zero values using advanced imputation methods (ALRA, scImpute, or MAGIC). The imputed data provides more accurate metabolic pathway activity calculations and feature selection in downstream analysis.

Note: This process is part of the ScrnaMetabolicLandscape process group. When using the full group, MetabolicExprImputation is automatically enabled (or skipped via noimpute flag). Use this skill when configuring the imputation step individually or when customizing imputation parameters within the ScrnaMetabolicLandscape group.

When to Use

• Second step in metabolic analysis workflow: After MetabolicInput, before MetabolicPathwayActivity and MetabolicFeatures
• High sparsity data: When your scRNA-seq data has many zero/dropout values (>50% zeros)
• Metabolic pathway sensitivity: When pathway activity calculations require complete expression matrices
• Before feature selection: When downstream MetabolicFeatures analysis needs imputed gene expression
• When imputation improves results: For datasets where dropout artifacts obscure biological signals

When to skip imputation:

• Data already imputed or complete (low dropout rate)
• Extremely large datasets where imputation is computationally expensive
• When you prefer to analyze raw unimputed expression

Configuration Structure

Process Enablement

As part of ScrnaMetabolicLandscape group (recommended):

[ScrnaMetabolicLandscape]
# Automatically includes MetabolicExprImputation

[ScrnaMetabolicLandscape.envs]
noimpute = false  # Set to true to skip imputation

Individual process configuration (advanced):

[ScrnaMetabolicLandscape.MetabolicExprImputation]
cache = true

Input Specification

MetabolicExprImputation automatically receives input from upstream processes:

• Input: Seurat object (RDS/qs format) from MetabolicInput
• Output: Imputed Seurat object with .imputed.qs suffix

Important: Input is automatically wired. No manual [ScrnaMetabolicLandscape.MetabolicExprImputation.in] specification needed.

Environment Variables

Core Settings

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
# Imputation method selection (required)
tool = "alra"  # Choice: "alra", "scimpute", "rmagic"

# Tool-specific configurations
alra_args = {}     # Type: json - Arguments for RunALRA()
scimpute_args = {} # Type: ns - Arguments for scImpute()
rmagic_args = {}   # Type: ns - Arguments for magic()

ALRA Configuration (Default - Fastest)

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "alra"
alra_args = {}  # Default: empty dict (use RunALRA() defaults)
# No additional parameters required

ALRA advantages:

• Fastest method (minutes vs hours)
• Low-rank approximation preserves global structure
• Zero-preserving imputation (no negative values)
• Built into Seurat (no external dependencies)

scImpute Configuration (Most Accurate)

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "scimpute"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
# Dropout threshold (genes with dropout rate > threshold will be imputed)
drop_thre = 0.5  # Type: float, Default: 0.5 (50% dropout)

# Number of clusters for imputation
kcluster = null  # Type: int or null, Default: null (auto-detect)

# Parallel cores
ncores = 1  # Type: int, Default: 1

# Reference gene file (optional)
refgene = ""  # Type: path, Default: "" (use all genes)

scImpute advantages:

• Cell-specific imputation (more accurate)
• Handles cell-type heterogeneity well
• Flexible clustering control
• Reference gene filtering option

scImpute considerations:

• Slower than ALRA (hours for large datasets)
• Higher memory usage
• Requires R package scImpute installation

MAGIC Configuration (Diffusion-Based)

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "rmagic"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.rmagic_args]
# Python interpreter path for magic-impute
python = "python"  # Type: path, Default: "python"

# Dropout threshold for gene selection
threshold = 0.5  # Type: float, Default: 0.5
# Only genes with dropout rates > threshold are imputed
# Dropout rate = (cells with non-zero expression) / (total cells)

MAGIC advantages:

• Diffusion-based approach preserves manifold structure
• Good for trajectory/continuum data
• Smooth expression patterns
• Handles complex data geometries

MAGIC considerations:

• Requires Python magic-impute package
• Intermediate speed (between ALRA and scImpute)
• Python path must be accessible
• May over-smooth sharp boundaries

Imputation Methods Comparison

| Method | Speed | Accuracy | Use Case | Dependencies | |--------|-------|----------|-----------|--------------| | ALRA | ⚡ Fast | ⭐⭐⭐ Good | Large datasets, quick analysis | Seurat (built-in) | | scImpute | 🐢 Slow | ⭐⭐⭐⭐ Best | Heterogeneous cell types, accuracy critical | R package scImpute | | MAGIC | 🚶 Medium | ⭐⭐⭐⭐ Good | Trajectory data, manifold preservation | Python magic-impute |

Method Selection Guide

Choose ALRA when:

• Dataset has >10,000 cells
• Computational speed is priority
• Data has clear cluster structure
• Memory resources are limited

Choose scImpute when:

• Dataset has <10,000 cells
• Accuracy is critical
• Cell-type heterogeneity is high
• You have sufficient compute resources

Choose MAGIC when:

• Data represents a continuum (e.g., differentiation)
• Manifold structure is important
• Trajectory analysis planned
• Python environment available

Configuration Examples

Minimal Configuration (Default ALRA)

[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "seurat_clusters"
# ALRA imputation enabled by default (noimpute = false)

Skip Imputation (Use Raw Data)

[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "seurat_clusters"
noimpute = true  # Skip MetabolicExprImputation entirely

scImpute with Custom Parameters

[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "Reactome_Pathways_2024"
group_by = "cluster"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "scimpute"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
drop_thre = 0.6   # Impute genes with >60% dropout
kcluster = 10       # Use 10 clusters
ncores = 4         # Parallelize with 4 cores

MAGIC for Trajectory Data

[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "pseudotime"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "rmagic"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.rmagic_args]
python = "/opt/conda/envs/r-base/bin/python"
threshold = 0.4  # Impute genes with >40% dropout

High-Performance Imputation (Large Dataset)

[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "seurat_clusters"
ncores = 8

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "alra"  # Fastest for large datasets

Conservative Imputation (Minimal Changes)

[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "scimpute"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
drop_thre = 0.8   # Only impute genes with >80% dropout (very sparse)
kcluster = 5        # Conservative clustering
refgene = "high_variance_genes.txt"  # Only impute specific genes

Common Patterns

Pattern 1: Standard Workflow (ALRA)

# Default fast imputation
[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "seurat_clusters"

Pattern 2: Sparse Data Handling (scImpute)

# For highly sparse data with cell-type heterogeneity
[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "Reactome_Pathways_2024"
group_by = "cluster"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "scimpute"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
drop_thre = 0.7  # High dropout threshold
ncores = 8         # More cores for speed

Pattern 3: Trajectory Analysis (MAGIC)

# For differentiation or developmental trajectories
[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "pseudotime"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "rmagic"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.rmagic_args]
threshold = 0.3  # Lower threshold for smoother gradients

Pattern 4: Large Dataset Optimization

# For >50k cells - prioritize speed
[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"
group_by = "seurat_clusters"
ncores = 16  # Use all available cores

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "alra"  # Only ALRA can handle this scale efficiently

Pattern 5: Benchmark Multiple Methods

# Compare imputation methods via cases
[ScrnaMetabolicLandscape]
[ScrnaMetabolicLandscape.envs]
gmtfile = "KEGG_2021_Human"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.cases.ALRA]
tool = "alra"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.cases.scImpute]
tool = "scimpute"

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.cases.scImpute.scimpute_args]
drop_thre = 0.5
ncores = 4

Dependencies

Upstream Processes

• Required: MetabolicInput (Seurat object from CombinedInput)
• CombinedInput sources:
  • SeuratClustering (most common)
  • TESSA (TCR-specific)
  • SeuratMap2Ref (reference mapping)
  • CellTypeAnnotation (cell type labels)

Downstream Processes (In ScrnaMetabolicLandscape Group)

• MetabolicPathwayActivity: Uses imputed expression for pathway scoring
• MetabolicPathwayHeterogeneity: Uses imputed expression for heterogeneity analysis
• MetabolicFeatures: Uses imputed expression for enrichment analysis

Package Dependencies

| Method | R Package | Python Package | Installation | |--------|-----------|----------------|--------------| | ALRA | alra (built-in to Seurat) | None | Included with Seurat | | scImpute | scImpute | None | install.packages("scImpute") | | MAGIC | Rmagic | magic-impute | pip install magic-impute |

Validation Rules

Expression Matrix Requirements

• Gene count: Minimum 1000 genes required for reliable imputation
• Cell count: scImpute requires >500 cells, ALRA/MAGIC work with any size
• Normalization: Seurat object must have normalized counts (typically LogNormalize or SCTransform)
• Assay availability: Default RNA assay must exist (imputed data stored here)

Method-Specific Validation

ALRA:

• No specific validation requirements
• Works with any Seurat object

scImpute:

• drop_thre: Must be between 0 and 1
• kcluster: If specified, must be > 0 and less than number of cells
• refgene: If provided, file must exist and contain valid gene names

MAGIC:

• python: Must point to valid Python interpreter with magic-impute installed
• threshold: Must be between 0 and 1
• Python environment must be accessible from R execution context

Output Validation

• Imputed assay: New assay RNA created with imputed data
• Original assay: Original assay renamed to RAW (with ALRA and MAGIC)
• No negative values: All imputed values should be non-negative
• Preserved dimensions: Gene and cell counts unchanged

Troubleshooting

Common Imputation Issues

Issue: Process runs indefinitely or very slowly

Symptoms: MetabolicExprImputation runs for >4 hours, especially with scImpute

Solutions:

1. Switch to ALRA (fastest method):

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs]
tool = "alra"

2. Reduce scImpute parameters:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
kcluster = 3        # Fewer clusters
ncores = 8          # More parallelization
drop_thre = 0.8     # Impute fewer genes

3. Skip imputation for large datasets:

[ScrnaMetabolicLandscape.envs]
noimpute = true

Issue: Memory errors during imputation

Symptoms: "Error: cannot allocate vector of size...", R session crashes

Solutions:

1. Use ALRA (lowest memory footprint)
2. Reduce ncores (less parallel memory):

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
ncores = 1

3. Filter genes before imputation:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
refgene = "highly_variable_genes.txt"

Issue: Negative values in imputed data

Symptoms: Downstream analysis fails or produces negative pathway scores

Causes:

• scImpute may produce negative values (expected behavior)
• MAGIC over-smoothing can create artifacts

Solutions:

1. Use ALRA (zero-preserving)
2. Set negative values to zero in downstream processes
3. Adjust threshold to be more conservative:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.rmagic_args]
threshold = 0.6  # Only impute very sparse genes

Issue: "Python not found" error with MAGIC

Symptoms: Error message about python or magic-impute not being found

Solutions:

1. Specify full Python path:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.rmagic_args]
python = "/opt/anaconda3/bin/python"

2. Install magic-impute in Python environment:

pip install magic-impute

3. Verify Python works from R:

system("python --version")

4. Switch to ALRA or scImpute if Python cannot be configured

Issue: Imputation doesn't improve results

Symptoms: Pathway activity scores similar before/after imputation

Solutions:

1. Check dropout rate - if low (<20%), imputation may not help
2. Try different method - ALRA vs scImpute vs MAGIC
3. Adjust threshold:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
drop_thre = 0.3  # Impute less sparse genes

4. Verify data quality - ensure QC was performed properly
5. Skip imputation if data quality is already good

Issue: Gene name mismatch after imputation

Symptoms: "Genes not found" errors in downstream processes

Solutions:

1. Check gene name format (case-sensitive):
   • Human: UPPERCASE (e.g., CD3D)
   • Mouse: TitleCase (e.g., Cd3d)
2. Verify GMT file matches Seurat object gene names
3. Use refgene to filter to valid genes only:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
refgene = "valid_genes.txt"

Issue: Imputed assay not accessible

Symptoms: Default assay still shows RAW data, downstream processes use wrong data

Causes:

• ALRA/MAGIC should automatically set imputed RNA as default
• Manual intervention or incorrect ordering

Solutions:

1. Verify default assay:

DefaultAssay(srtobj)  # Should return "RNA", not "RAW"

2. Manually set default assay (if needed):

DefaultAssay(srtobj) <- "RNA"

3. Check output file - ensure .imputed.qs was created

Issue: scImpute clustering fails

Symptoms: Error about kcluster or clustering algorithm

Solutions:

1. Set kcluster to null (auto-detect):

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
kcluster = null

2. Reduce kcluster if manual:

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
kcluster = 5

3. Switch to ALRA if clustering issues persist

Issue: Imputation creates artifacts

Symptoms: Unusual expression patterns, biologically implausible values

Causes:

• Over-imputation (threshold too low)
• Wrong imputation method for data type
• Poor quality input data

Solutions:

1. Increase threshold (impute fewer genes):

[ScrnaMetabolicLandscape.MetabolicExprImputation.envs.scimpute_args]
drop_thre = 0.8  # Only impute extremely sparse genes

2. Try different method:

   • Use ALRA for cluster-based data
   • Use MAGIC for trajectory data
   • Use scImpute for heterogeneous data

3. Improve QC upstream before imputation

4. Skip imputation if artifacts severe

External References

Original Papers

ALRA:

• Linderman, G. C., Zhao, J., & Kluger, Y. (2018). Zero-preserving imputation of scRNA-seq data using low-rank approximation. Nature Communications, 12(1), 6375. https://www.nature.com/articles/s41467-021-27729-z

scImpute:

• Li, W. V., & Li, J. J. (2018). An accurate and robust imputation method scImpute for single-cell RNA-seq data. Nature Communications, 9(1), 997. https://www.nature.com/articles/s41467-018-03405-7

MAGIC:

• Van Dijk, D., et al. (2018). Recovering Gene Interactions from Single-Cell Data Using MAGIC. Cell, 174(3), 716-729. https://www.cell.com/cell/fulltext/S0092-8674(18)30724-4

Metabolic Analysis Framework

• Original paper: Xiao, Z. et al. "Metabolic landscape of tumor microenvironment at single cell resolution." Nature Communications 10, 1-12 (2019) https://www.nature.com/articles/s41467-019-11738-0
• GitHub repository: https://github.com/LocasaleLab/Single-Cell-Metabolic-Landscape
• biopipen documentation: https://pwwang.github.io/biopipen/pipelines/scrna_metabolic/

Tool Documentation

• Seurat ALRA: https://satijalab.org/seurat/reference/runalra
• scImpute: https://github.com/vvnathan/scImpute
• MAGIC: https://github.com/KrishnaswamyLab/MAGIC
• magic-impute: https://pypi.org/project/magic-impute/

Package Installation

# ALRA (Seurat built-in)
install.packages("Seurat")

# scImpute
install.packages("scImpute")

# Rmagic (for MAGIC)
install.packages("Rmagic")

# magic-impute (Python)
pip install magic-impute

Decision Tree for Imputation Method Selection

Start: MetabolicExprImputation
│
├─ Dataset size > 50k cells?
│  └─ YES → Use ALRA (tool = "alra")
│
├─ Dataset size 10k-50k cells?
│  ├─ Priority: Speed?
│  │  └─ YES → Use ALRA
│  └─ Priority: Accuracy?
│     └─ YES → Use scImpute (tool = "scimpute")
│
├─ Dataset size < 10k cells?
│  ├─ Cell-type heterogeneity high?
│  │  └─ YES → Use scImpute
│  └─ Continuum/trajectory data?
│     └─ YES → Use MAGIC (tool = "rmagic")
│
└─ Data quality concerns?
   ├─ Low dropout rate (<20%)?
   │  └─ Skip imputation (noimpute = true)
   └─ High dropout rate (>80%)?
      └─ Use scImpute with high threshold (drop_thre = 0.8)

Performance Benchmarks

| Dataset Size | ALRA | scImpute (4 cores) | MAGIC | |--------------|-------|-------------------|-------| | 1,000 cells | 2 min | 5 min | 3 min | | 5,000 cells | 5 min | 15 min | 8 min | | 10,000 cells | 10 min | 45 min | 15 min | | 25,000 cells | 25 min | 2-3 hours | 45 min | | 50,000 cells | 1 hour | 8-10 hours | 2 hours | | 100,000 cells | 2 hours | Not recommended | 4 hours |

Note: Benchmarks on 16-core machine, 64GB RAM. scImpute scales quadratically with cell count.