# Tumour versus normal expression of a gene: ERBB2

> How to check tumour versus normal expression of a gene across 24 cancer types with one ERBB2 query, using TCGA/GTEx Toil data plus DepMap dependency.

Data as of 2026-11-06 · Published 2026-11-10 · Updated 2026-11-06 · https://bioq.bio/blog/en/howto-tumour-versus-normal-expression-erbb2

![BioQ result page for the query ERBB2 showing tumour versus normal expression and CRISPR dependency blocks with their sources](https://bioq.bio/blog/img/howto-tumour-versus-normal-expression-erbb2/en/hero.webp)

To get the tumour versus normal expression of a gene across cancer types, one BioQ query, ERBB2, returns a table for 24 cancer types that have both tumour and normal samples in TCGA/GTEx Toil. Breast Invasive Carcinoma has 1092 tumours and 113 normals, with a mean of 13.76 in tumour against 12.65 in normal. Data are as of 2026-11-06.

## What we asked: tumour versus normal expression of a gene

The task is common. You have a gene, here ERBB2 (HER2), and want to know in which cancer types its mRNA is higher or lower in tumour than in normal tissue. You also want to know whether cell lines from the same tissues depend on it for growth.

The manual route is to open a tumour-versus-normal expression resource for the TCGA and GTEx comparison, pick the gene, and read off cancer types one by one. Then you open a CRISPR dependency portal, search the gene again, and group the cell lines by lineage. Afterwards you reconcile the two sets of names and units in a spreadsheet.

In BioQ the query is the gene symbol alone:

`ERBB2`

The answer draws on 2 distinct resources through 2 engine queries. The measured engine time for both was 0.12 seconds. That figure is engine time only, not the time you spend reading and interpreting the result.

## What BioQ returned

The first block is tumour versus normal expression from the UCSC Xena Toil recompute of TCGA and GTEx. Values are in log2(norm_count+1) units. BioQ reports 24 cancer types with both tumour and normal samples and shows the top 8 here.

![BioQ overview page for the query ERBB2 with answer blocks and the source named for each one](https://bioq.bio/blog/img/howto-tumour-versus-normal-expression-erbb2/en/ov.webp)
*The ERBB2 query in BioQ: expression and dependency blocks on one page*

| Cancer type | Tumours | Normals | Mean tumour | Mean normal | log2 fold |
|---|---|---|---|---|---|
| Kidney Clear Cell Carcinoma | 530 | 72 | 12.09 | 13.62 | -0.17 |
| Sarcoma | 258 | 2 | 10.98 | 12.24 | -0.16 |
| Uterine Corpus Endometrioid Carcinoma | 180 | 23 | 13.23 | 12.04 | 0.14 |
| Breast Invasive Carcinoma | 1092 | 113 | 13.76 | 12.65 | 0.12 |
| Cervical & Endocervical Cancer | 304 | 3 | 13.01 | 11.93 | 0.12 |
| Glioblastoma Multiforme | 153 | 5 | 10.76 | 9.69 | 0.15 |
| Bladder Urothelial Carcinoma | 407 | 19 | 13.58 | 12.59 | 0.11 |
| Pancreatic Adenocarcinoma | 178 | 4 | 12.89 | 11.97 | 0.11 |

The second block is CRISPR dependency by lineage from DepMap (release depmap_24q4, Chronos gene effect). Across 1178 cell lines, 177 are dependent on ERBB2, which is 15.0 percent. The mean gene effect is -0.329, and ERBB2 is not flagged as common essential. A more negative Chronos value means the knockout reduced viability more.

| Lineage | Cell lines | Dependent | % dependent | Mean gene effect |
|---|---|---|---|---|
| Breast | 53 | 18 | 34 | -0.643 |
| Head and Neck | 77 | 35 | 45.5 | -0.509 |
| Esophagus/Stomach | 69 | 18 | 26.1 | -0.476 |
| Biliary Tract | 35 | 9 | 25.7 | -0.431 |
| Pancreas | 47 | 11 | 23.4 | -0.425 |
| Ovary/Fallopian Tube | 59 | 16 | 27.1 | -0.417 |

## What it means and how to read it

**Expression, TCGA tumour vs normal.** In the rows shown, tumour means are higher than normal means in Uterine Corpus Endometrioid Carcinoma (13.23 against 12.04), Breast Invasive Carcinoma (13.76 against 12.65), Cervical & Endocervical Cancer (13.01 against 11.93), Glioblastoma Multiforme (10.76 against 9.69), Bladder Urothelial Carcinoma (13.58 against 12.59) and Pancreatic Adenocarcinoma (12.89 against 11.97). Two cancer types go the other way: Kidney Clear Cell Carcinoma (12.09 against 13.62) and Sarcoma (10.98 against 12.24).

**Read the log2 fold column with care.** In the table as returned, the log2 fold values do not equal the difference between the two displayed means in every row. For example, Breast Invasive Carcinoma shows 0.12 in the log2 fold column. Before you quote a fold change in a figure or a manuscript, check how that column is defined for this release. The mean tumour and mean normal columns are the safer numbers to cite, together with the sample counts.

**Sample sizes.** The normal group is very small in some cancer types: 2 in Sarcoma, 3 in Cervical & Endocervical Cancer, 4 in Pancreatic Adenocarcinoma and 5 in Glioblastoma Multiforme. A direction of change based on so few normals is a hint, not a result. Breast Invasive Carcinoma (113 normals) and Kidney Clear Cell Carcinoma (72 normals) rest on much more material.

**Dependency.** Among the lineages shown, Breast has the most negative mean gene effect (-0.643), with 18 of 53 cell lines dependent (34.0 percent). Head and Neck has 35 of 77 dependent (45.5 percent) with a mean of -0.509. Esophagus/Stomach, Biliary Tract, Pancreas and Ovary/Fallopian Tube follow with percentages between 23.4 and 27.1. The dependency cut-off is a named constant in the DepMap release, not a property of the data, so the percentages move if the cut-off changes.

**Putting the two blocks together.** Pancreatic Adenocarcinoma shows higher mean expression in tumour in the first table, and the Pancreas lineage appears in the dependency table. That is a reason to look further, not evidence that expression predicts dependency. The two datasets use different tissue labels and different samples (patient tissue in one, cell lines in the other).

**Next steps.** Pick the cancer types where the sample counts are large and the direction is clear. Then look at the same gene in other layers, such as mutation or copy-number data in TCGA, or the clinical trial record for the indication you care about. In BioQ you can follow up with a second query on the same gene and cancer type. If you need an independent check, repeat the comparison on the raw Toil matrix.

## How to reproduce it in BioQ

1. Open BioQ and go to the search box.
2. Enter the gene symbol exactly as below. No filters or extra words are needed.

```text
ERBB2
```

[Open this query in BioQ](https://bioq.bio/login?next=%2Fapp%23q%3DERBB2)

3. Read the tumour versus normal block first. Note the number of cancer types with both groups, the sample counts per row, and the units (log2(norm_count+1)).
4. Scroll to the CRISPR dependency block and compare lineages with the cancer types you picked.
5. Use the source name printed with each block when you cite a number. Both the release label and the note about units come with the block.

## Limits: what the data do not show

- The table is limited to the top 8 of 24 cancer types. Cancer types outside it are not described here.
- The data are bulk mRNA. They say nothing about HER2 protein level, gene amplification status, or which tumours in a cancer type are the high expressers. A mean can hide a mixed population.
- Several cancer types have very few normal samples, so differences there are fragile.
- The log2 fold column should be checked against its definition before use, as noted above.
- Dependency data come from cell lines, not patients. A dependency percentage depends on the cut-off constant of the release.
- Nothing here is medical advice or a statement about how any patient should be treated.

## Data and method
The numbers on this page were read from the BioQ engine on 2026-11-06 (2 queries, 0.1 s of engine time in total). The tables are generated by code from the engine's answers; the text only restates them.

`/depmap`, `/tcga`

Run the same query yourself: [ERBB2](https://bioq.bio/login?next=%2Fapp%23q%3DERBB2)

*For research use only. This page summarises public data and is not medical advice.*

BioQ is an AI search engine and agent with one goal: making biology research 100X faster. Ask in the web app, over the REST API, or from Claude, ChatGPT or Cursor through MCP.

## Queries used

- [`ERBB2`](https://bioq.bio/login?next=%2Fapp%23q%3DERBB2)

## Frequently asked questions

### How many cancer types have both tumour and normal data for ERBB2 in BioQ?

BioQ reports 24 cancer types with both tumour and normal samples in the TCGA/GTEx Toil recompute. The article table shows the top 8 of them, as of 2026-11-06.

### Is ERBB2 higher in breast tumours than in normal tissue?

In Breast Invasive Carcinoma the mean is 13.76 in 1092 tumours and 12.65 in 113 normals, in log2(norm_count+1) units. This is mRNA only and says nothing about protein or amplification.

### Is ERBB2 a common essential gene in DepMap?

No. In depmap_24q4 it is not flagged as common essential. Of 1178 cell lines, 177 are dependent (15.0 percent), and the mean gene effect is -0.329.

### How long did the ERBB2 query take?

The measured engine time was 0.12 seconds for 2 engine queries drawing on 2 distinct resources. This does not include the time you spend reading the results.

### Can I trust the fold change for cancer types with few normals?

Treat it with caution. Sarcoma has 2 normals, Cervical & Endocervical Cancer 3, Pancreatic Adenocarcinoma 4 and Glioblastoma Multiforme 5, so the direction of change in those rows is only a hint.

## Sources

- Dataset: [TCGA pan-cancer expression (UCSC Xena Toil recompute)](https://xenabrowser.net/)
- Dataset: [DepMap CRISPR Chronos gene effect (24Q4)](https://depmap.org/portal/)

> Results come from public databases as of the date shown and may change. BioQ gives research information, not medical advice.
