Autoannotation

Auto annotation with Nygen Insights

Auto annotations will create an AI report on cell type suggestions for each cluster of your chosen group. This feature uses the marker genes from each cluster to predict their cell types.

💡 Note: Nygen Insights is ScarfWeb's free LLM-augmented cell type annotation feature. The steps on this page walk through Nygen Insights.

Key features

Under the hood, the auto annotation functionality integrates four external canonical marker databases:

• Human Commons Cell Atlas: A comprehensive resource for human cell type markers
• ACT: Animal Cell Type Atlas, providing markers across multiple species
• PanglaoDB: A database for single-cell RNA sequencing experiments from mouse and human
• CellMarker2: An updated version of the CellMarker database with expanded cell type information

These databases are consumed to generate initial cluster annotations based on the integrated databases. This, along with Cell Ontology database provides a really good context to LLMs for predicting granular cell types with high confidence.

Learn more: Go further with CyteType in ScarfWeb

CyteType is now natively integrated into ScarfWeb, so biologists interpreting single-cell data can use multi-agent characterization that was previously limited to computational biology teams. You can go a step beyond reliable Nygen Insights annotations when you need deeper cluster characterization. CyteType has been benchmarked against leading methods, with accuracy improvements of 388% over GPTCellType, 268% over CellTypist, and 101% over SingleR.

Reannotate any dataset in the workspace and get more than labels alone. Each cluster provides:

• Evidence behind its identity call
• Disease context and intercellular signaling associated with it
• An interactive report that allows deeper exploration of the biology with our research agent

CyteType annotation in ScarfWeb is rate limited.

Creating a new auto annotation report

Steps

1. From the explorer page of your dataset, you can generate auto annotation reports from the Auto Annotate button with the list of clusters in the bottom right corner.

2. Click on Create new annotation report.

3. Fill in the rest of the form for submission with details of your dataset.

   • Required fields
     • Select annotation species The report will be generated based on marker genes, choose the species that matches best with your dataset.
     • Select group Select from a list of the current groups or clusters to annotate.
     • Additional information Providing more details on your dataset will give the LLM more context, this can improve the results and provide more relevant reports

4. The report will have a few minutes to generate. Once it’s done, you can access all your generated reports from the Nygen Insights panel. More on report content below.

Report features

Auto annotations not only suggest cell type annotations, but also provide detailed supporting information, including marker genes that support or conflict with each cell type, confidence levels, and additional metrics. This automated assistance streamlines the manual curation process, allowing researchers to dedicate more time to focused investigations of rare cell types, subtle cellular states, and other detailed analyses.

Details

Each report will include the input parameters used which will help keep track of the context given to generate the report.

Each cluster of the selected group will be given a suggested cell type annotation. Below the cluster name, the coloured bar is given to indicate the confidence level of the predicted cell type.

Cell Ontology

This will open a new tab to the Ontology Lookup Service website page, where you can find the cell ontology details on the suggested cell type.

Use name

You can rename the cluster using the new cell annotation.

The report includes text details on what supported the cell annotations. This includes supporting marker genes in the cluster, as well as conflicting markers and anomalies which may lower the confidence of the auto annotation.

Alternative cell types to the current annotation and similar clusters are also suggested.