scRNA-seq Troubleshooting Guide 2025

Solutions:

1. Adjust Quality Thresholds:

   # R/Seurat example
   # Lower thresholds for difficult samples
   subset(seurat_obj, subset = nFeature_RNA > 500 & nCount_RNA > 1000)
   
   # Python/Scanpy example  
   sc.pp.filter_cells(adata, min_genes=500)
   sc.pp.filter_genes(adata, min_cells=10)

2. Check Sample Preparation: Review tissue dissociation, cell capture efficiency
3. Consider Cell Type: Some cell types naturally have lower RNA content

Solutions:

1. Progressive Filtering:

   # Start with relaxed thresholds
   subset(seurat_obj, subset = percent.mt < 25)
   # Then tighten based on results
   subset(seurat_obj, subset = percent.mt < 15)

2. Tissue-Specific Thresholds: Brain cells: <20%, Blood cells: <10%
3. Regression Approach: Regress out mitochondrial effects instead of filtering

Solutions:

1. Try Different Normalization Methods:

   Method	Best For	When to Use
   Log-normalization	Standard analysis	Default choice
   SCTransform	Heterogeneous datasets	Strong batch effects
   scran	Sparse data	Many zero counts

2. Parameter Optimization: Adjust scaling factors and regression variables
3. Alternative Approaches: Consider SCTransform for complex datasets

Solutions:

1. Increase HVG Count: Try 3000-5000 highly variable genes instead of 2000
2. Use Multiple Methods: Combine vst, mean.var.plot, and dispersion approaches
3. Manual Curation: Remove unwanted gene categories (ribosomal, mitochondrial)

Solutions:

1. Reduce Resolution:

   # Try lower resolution values
   FindClusters(seurat_obj, resolution = 0.3)  # Instead of 0.8
   FindClusters(seurat_obj, resolution = 0.5)  # Middle ground

2. Increase k Parameter: Use more neighbors in SNN graph construction
3. Merge Similar Clusters: Use hierarchical clustering to identify merge candidates

Solutions:

1. Increase Resolution: Try 0.8, 1.0, or higher
2. Adjust PCA Dimensions: Use more PCs (30-50 instead of 20)
3. Re-examine Preprocessing: Check if important genes were filtered out

Solutions with mLLMCelltype:

1. Enable Multi-Model Consensus:

   🎯 Pro Tip: Use 3-5 different AI models and set consensus threshold to 0.7 for higher confidence

2. Use Discussion Mode: Let models debate uncertain annotations
3. Provide Better Context: Include tissue type and experimental condition information
4. Refine Marker Genes: Use more specific, high-quality marker genes

Solutions:

1. Lower Consensus Threshold: Allow more exploratory annotations
2. Manual Review: Examine clusters with "Unknown" annotations
3. Literature Search: Research potential novel populations in your tissue
4. Functional Analysis: Perform pathway analysis to understand cell function

Integration Method Comparison:

Solutions:

1. Reduce Data Size: Subsample cells or genes for initial analysis
2. Use Disk-Based Storage: Enable on-disk storage for large objects
3. Optimize Parameters: Reduce PCA dimensions, use fewer HVGs
4. Cloud Computing: Use mLLMCelltype web platform for processing