You: Search PubMed for recent papers on single-cell RNA-seq analysis
     of pancreatic islet cells

[Supervisor → Research Agent]
Searching PubMed for "single-cell RNA-seq pancreatic islet cells"...

Found 5 recent papers (2024-2026):

1. "Single-cell atlas of human pancreatic islets reveals heterogeneity
    in beta cell dysfunction"
   PMID: 39123456 | Nature Communications | Dec 2025
   Methods: 10X Genomics, Seurat v5, UMAP clustering
   GEO: GSE198765

2. "Trajectory analysis of alpha-to-beta cell transdifferentiation"
   PMID: 39234567 | Cell Reports | Nov 2025
   Methods: Smart-seq2, Monocle3 pseudotime
   ...

You: Download the GEO dataset from the first paper

[Supervisor → Data Expert Agent]
Downloading GSE198765 from GEO...
- Series: GSE198765
- Samples: 12 (4 control, 4 T1D, 4 T2D)
- Platform: 10X Genomics 3' v3
- Cells: ~45,000

Download complete. Data loaded as 'gse198765.h5ad'

You: Load my dataset pancreatic_cells.h5ad and run QC

[Supervisor → Transcriptomics Expert]
Loading pancreatic_cells.h5ad...

Dataset summary:
- Cells: 15,234
- Genes: 32,456
- Batches: 3

Running QC metrics...

QC Results:
| Metric | Mean | Median | Threshold |
|--------|------|--------|-----------|
| nGenes | 2,456 | 2,312 | >500 |
| nUMIs  | 8,923 | 7,654 | >1000 |
| %Mito  | 4.2%  | 3.8%  | <10% |

Cells flagged for removal: 423 (2.8%)
- Low gene count: 156
- High mitochondrial: 267

Shall I filter these cells and proceed to normalization?

You: Yes, filter and normalize

[Transcriptomics Expert]
Filtering 423 low-quality cells...
Remaining: 14,811 cells

Normalizing with scran pooling method...
Log-transforming...

Normalization complete. Ready for:
- Highly variable gene selection
- Dimensionality reduction
- Clustering

You: Find clusters and annotate cell types

[Transcriptomics Expert → Annotation Expert]
Selecting 2,000 highly variable genes...
Running PCA (50 components)...
Building neighbor graph (k=15)...
Clustering with Leiden algorithm (resolution=0.8)...

Found 12 clusters:

| Cluster | Cells | Top Markers | Predicted Type |
|---------|-------|-------------|----------------|
| 0 | 3,456 | INS, IAPP, HADH | Beta cells |
| 1 | 2,134 | GCG, ARX, IRX2 | Alpha cells |
| 2 | 1,876 | SST, HHEX | Delta cells |
| 3 | 1,234 | PPY, SERTM1 | PP cells |
| 4 | 987 | PRSS1, CTRB2 | Acinar cells |
| ... | ... | ... | ... |

[Generating UMAP visualization]

Annotation confidence: High (all clusters matched known markers)

You: Run differential expression between diabetic and control samples

[Supervisor → DE Analysis Expert]
Detecting sample metadata...
Found condition column: 'disease_status'
- Control: 4,234 cells
- Diabetic: 4,567 cells

Running DE analysis (Wilcoxon rank-sum test)...
Multiple testing correction: Benjamini-Hochberg

Significant genes (adjusted p < 0.05, |log2FC| > 1):

Upregulated in Diabetic (234 genes):
| Gene | log2FC | adj_pval | Description |
|------|--------|----------|-------------|
| TXNIP | 2.45 | 1.2e-45 | Oxidative stress |
| DDIT3 | 1.89 | 3.4e-32 | ER stress |
| ATF4 | 1.67 | 5.6e-28 | UPR pathway |

Downregulated in Diabetic (156 genes):
| Gene | log2FC | adj_pval | Description |
|------|--------|----------|-------------|
| INS | -2.12 | 2.1e-67 | Insulin |
| MAFA | -1.78 | 4.5e-34 | Beta cell TF |
| PDX1 | -1.45 | 8.9e-28 | Beta cell TF |

[Generating volcano plot]

Shall I run pathway enrichment on these gene lists?

You: Run pathway enrichment on the upregulated genes

[DE Analysis Expert]
Running Gene Ontology enrichment...

GO Biological Process (top 10):
| GO Term | Genes | p-value | Description |
|---------|-------|---------|-------------|
| GO:0034976 | 45 | 1.2e-23 | Response to ER stress |
| GO:0006986 | 38 | 3.4e-19 | Response to unfolded protein |
| GO:0070059 | 32 | 5.6e-17 | Intrinsic apoptotic signaling |

KEGG Pathways:
| Pathway | Genes | p-value |
|---------|-------|---------|
| hsa04141 | 28 | 2.3e-15 | Protein processing in ER |
| hsa04210 | 24 | 4.5e-12 | Apoptosis |
| hsa04932 | 21 | 7.8e-10 | Non-alcoholic fatty liver |

[Generating enrichment dot plot]

The results suggest ER stress and UPR activation are major features
of diabetic beta cells.

You: I uploaded spectronaut_export.tsv. Can you process it?

[Supervisor → Proteomics Expert]
Detecting file format... Spectronaut report detected.

Parsing Spectronaut export...
- Proteins: 4,567
- Peptides: 34,567
- Samples: 24 (12 treatment, 12 control)

Running QC...
- Missing values: 8.3%
- CV distribution: median 12.4%
- Sample correlations: 0.89-0.97

Shall I proceed with:
1. Imputation (MinProb or KNN)
2. Normalization (median or quantile)
3. Differential analysis

You: Use KNN imputation, median normalization, then run DE

[Proteomics Expert]
Imputing with KNN (k=5)... Complete.
Normalizing with median centering... Complete.

Running limma-based differential analysis...

Significant proteins (adj.p < 0.05, |log2FC| > 1): 234

Top hits:
| Protein | Gene | log2FC | adj_pval |
|---------|------|--------|----------|
| P04637 | TP53 | 2.34 | 1.2e-12 |
| Q9Y6K9 | NEMO | 1.89 | 3.4e-10 |
| ... | ... | ... | ... |

[Generating protein volcano plot]

You: I have both RNA-seq and proteomics from the same samples.
     Can you integrate them?

[Supervisor → ML Expert]
Loading datasets...
- RNA-seq: 12,456 genes x 24 samples
- Proteomics: 4,567 proteins x 24 samples

Matching samples by ID... 24/24 matched.
Matching features by gene symbol... 3,234 genes/proteins matched.

Running MOFA+ multi-omics factor analysis...
- Factors: 10
- Variance explained: 78.4% (RNA), 65.2% (Proteomics)

Factor interpretation:
| Factor | RNA var% | Prot var% | Top genes | Interpretation |
|--------|----------|-----------|-----------|----------------|
| F1 | 23.4% | 18.9% | TP53, MDM2, CDKN1A | p53 pathway |
| F2 | 15.6% | 12.3% | HIF1A, VEGFA, EGLN1 | Hypoxia response |
| F3 | 11.2% | 14.5% | MYC, CCND1, CDK4 | Cell cycle |

[Generating factor correlation heatmap]

Good: "I have 10X Genomics single-cell data from mouse liver.
       There are 3 conditions: control, acute injury, and chronic injury.
       I want to identify cell types affected by injury."

Less good: "Analyze my data"

"Why did you choose the Leiden algorithm over Louvain?"
"What does this pathway enrichment tell us biologically?"
"Can you explain what the UMAP coordinates represent?"