1 of 24

Single-Cell RNA-Seq Analysis of Microglia in Alzheimer’s Disease

Ifthekar Hussain�Computational Analysis Task

2 of 24

Study Objectives

Identify microglial gene expression changes across disease groups:

NND → nAD → iAD

Compare:

NND (healthy) vs nAD (Alzheimer’s)
nAD vs iAD (immunized Alzheimer’s)�

Identify key microglial genes that are significantly up- or down-regulated�
Perform GO Biological Process enrichment to link genes to pathways and functions

3 of 24

Dataset & Study Design

Dataset:

GSE263034, scRNA-seq
25 samples
Donors: NND = 6, nAD = 6, iAD= 13

Analysis Workflow

4 of 24

Quality Control ( Before Filtering)

Dataset contains variable cell quality
Outliers with:

Very low features
Very low UMIs
High mitochondrial percentage

Applied QC thresholds to remove low-quality

5 of 24

.

nCount vs nFeature:�In this plot, I’m checking whether each cell has a healthy transcriptome. Good cells follow this smooth increasing curve, high UMIs, high gene complexity. The scattered low-feature cells underneath the curve are classic low-quality droplets, so this helped me define my lower gene/UMI thresholds.

percent.mt vs nCount:

Here I’m looking for stressed or dying cells. Cells with high mitochondrial percent but low RNA complexity cluster on the left side. That pattern is exactly where mitochondrial leakage shows up biologically, so this plot guided my percent.mt cutoff.

percent.mt vs nFeature:

This plot is just a double-check. Good cells have low mitochondrial % and lots of genes. If a cell has high mito levels and very few genes, it usually means the cell is unhealthy, so I removed those to keep only good-quality microglia.

6 of 24

Quality Control(After Filtering)

Removed cells with extremely low RNA content

nCount_RNA < (median − 3 × MAD) per-sample threshold
nFeature_RNA < (median − 2 × MAD) per-sample threshold

Removed high-mitochondrial cells

percent.mt > 20%

Removed doublets

Detected using scDblFinder
Filtered separately for each sample

7 of 24

Doublet Detection

Doublets identified using scDblFinder (sample - wise detection)
Total cells after QC filtering : 112,196
Total cells after doublet removal: 106,298
Doublets removed: 5,898 cells
Final retention: 94.7% singlets

8 of 24

.

9 of 24

Normalization(SCTransform)

Normalized all 25 samples using SCTransform
Regressed mitochondrial percentage(percent.mt)
Removed technical noise and stabilized variance
Result: 106,298 high-quality cells ready for integration

10 of 24

Batch Correction(Harmony Integration)

Combined all 25 SCTransformed samples for integration
Performed PCA to capture major variance components
Selected 30 PCs based on the elbow plot
Applied Harmony to correct donor specific batch effects
Samples align well after correction while preserving true biology

Selected 30PCs

11 of 24

Before Integration (Batch Effects)

UMAP shows strong batch effects before integration
Samples cluster by donor, not biology
Technical variation dominates the structure, hiding true biology
Confirms the need for batch correction (Harmony)

12 of 24

After Integration(Harmony Corrected)

Harmony successfully removed donor-specific batch effects
Samples from all donors now overlap instead of forming separate clusters
Technical variation is greatly reduced
UMAP structure is cleaner and more biologically meaningful
Clustering becomes more stable after correction

13 of 24

Cluster Marker Identification (FindAllMarkers)

Identified top marker genes for each cluster using SCT-normalized data
Applied standard cutoffs: log2FC > 0.25 and min.25% expression
Extracted top 10 genes per cluster
Marker profiles used for biological interpretation
Basis for accurate cell-type annotation

14 of 24

Marker Gene Heatmap

15 of 24

Cluster Annotation(cell-type identification)

Identified top marker genes for each cluster
Compared marker profiles with established cell-type signatures
Assigned biological labels to clusters based on marker expression patterns
Added final labels to metadata and visualized on UMAP

Microglia clusters

16 of 24

Microglia Subsetting & Re-clustering

Extracted microglia by selecting clusters 20, 21 & 23 from the Harmony-integrated UMAP
Confirmed microglia identity using average expression of canonical markers
Subset contained 2,133 microglial cells
Re-processed microglia subset using:

Log Normalize
FindVariableFeatures
ScaleData
PCA

Re-clustered microglia subset

PC15 = Inflection point

17 of 24

Microglia Reclustering & Marker Based Cell Type Annotation

Re-clustered microglia subset using 15 PCs and resolution = 0.3
Identified top markers with FindAllMarkers
Annotated clusters using canonical marker genes
Final UMAP shows clear separation of microglial states

18 of 24

.

19 of 24

Differential Expression Analysis(MAST) on Microglia

Subsetted microglial states only (Homeostatic, Stress, Inflammatory, DAM-like MG)
Performed differential expression using MAST (single-cell hurdle model)
Two comparisons:

nAD vs NND
iAD vs nAD

Adjusted p-values using Benjamini–Hochberg (FDR < 0.05)
Extracted top up- and down-regulated genes for each comparison
Visualized significant genes using volcano plots (log2FC > 0.585, FDR < 0.05) to show up-regulated, down-regulated, and non-significant genes

20 of 24

Volcano Plot - nAD vs NND

21 of 24

Volcano plot - iAD vs nAD

22 of 24

GO Enrichment- nAD vs NND

.

23 of 24

GO Enrichment- iAD vs nAD

.

24 of 24

Conclusions

nAD microglia show significant metabolic stress and reduced neuroprotective function
iAD microglia show restored protective phenotype with decreased stress markers
Amyloid-β immunization shifts microglial state toward neuroprotection
Results suggest immunotherapy improves microglial health in Alzheimer's disease