Integration - Answer Key

Author

Noor Sohail

Published

April 19, 2026

Exercise 1

Do you see a clear split based on sample in the UMAP and PCA? What could this indicate about the presence of batch effects in our data?

Figure 1: PCA and UMAP colored by sample.

When we look at the UMAP and PCA plot without any other pieces of information it does seem like there is a batch effect. In the PCA space, we can see in the top left-hand corner of PC space that the bins predominantly belong to P5CRC. Similarly in the UMAP space, there is a cluster of bins on the right-hand side that belong solely to P5CRC and another at the top of the UMAP.

Based solely on this visualization and no additional information, it may seem like integration may be necessary. However, the other bins in the dataset do appear to overlay quite well across both datasets.

Exercise 2

Is there a biological reason we see clusters that are dominated by P5CRC bins? What cell type do you think these clusters may represent?

When we look at the celltypes we anticipate being in our dataset, there is one population that stands outs - Tumor cells.

The P5NAT sample is “normal adjacent tissue” which indicates that we should not see any tumor cells in this sample. Conversely, we are anticipating cancerous cells in the P5CRC dataset. So before we make a final decision on integration, we should first see if the previously identified populations could possible be tumor cells.

Back to Lesson >>

Back to Schedule

Reuse

CC-BY-4.0

---
title: "Integration - Answer Key"
author:
  - Noor Sohail
date: "2026-04-19"
license: "CC-BY-4.0"
editor_options: 
  markdown: 
    wrap: 72
---

```{r}
#| label: load_libraries_data
#| echo: false
# Load libraries and data
library(Seurat)
library(tidyverse)
seurat_processed <- qs2::qs_read("intermediate/08_seurat_processed.qs")
```

# Exercise 1

1. Do you see a clear split based on sample in the UMAP and PCA? What could this indicate about the presence of batch effects in our data?

```{r}
#| label: fig-pca_umap_by_sample
#| fig-cap: PCA and UMAP colored by sample.
#| fig-width: 10
#| echo: false
p1 <- DimPlot(seurat_processed,
              group.by="orig.ident",
              reduction = "full.pca.sketch") +
  NoLegend() + ggtitle("PCA (Projected)")

p2 <- DimPlot(seurat_processed,
              group.by = "orig.ident",
              reduction = "full.umap.sketch") +
  ggtitle("UMAP (Projected)")

p1 | p2
```

When we look at the UMAP and PCA plot without any other pieces of information it does seem like there is a batch effect. In the PCA space, we can see in the top left-hand corner of PC space that the bins predominantly belong to `P5CRC`. Similarly in the UMAP space, there is a cluster of bins on the right-hand side that belong solely to `P5CRC` and another at the top of the UMAP.

Based solely on this visualization and no additional information, it may seem like integration may be necessary. However, the other bins in the dataset do appear to overlay quite well across both datasets.

# Exercise 2

2. Is there a biological reason we see clusters that are dominated by `P5CRC` bins? What cell type do you think these clusters may represent?

When we look at the celltypes we anticipate being in our dataset, there is one population that stands outs - `Tumor cells`.

The `P5NAT` sample is "normal adjacent tissue" which indicates that we should not see any tumor cells in this sample. Conversely, we are anticipating cancerous cells in the `P5CRC` dataset. So before we make a final decision on integration, we should first see if the previously identified populations could possible be tumor cells.

***

[Back to Lesson >>](09_integration.qmd)

[Back to Schedule](../schedule/schedule.qmd)