Skip to content

JBrowse chat bot #2441

@ValWood

Description

@ValWood

We could try this. It came up on the AI workshop yesterday. These are some questions I asked about its capabilities.

what sort of questions can the chatbot plugin for JBrowse called jbrowse-plugin-assistant-chatbot. answer
ChatGPT said:

The jbrowse-plugin-assistant-chatbot can handle questions related to genome browsing and data exploration in JBrowse 2, including:

Navigation & UI help

How to zoom, scroll, add/remove tracks, configure views, or use specific tools in JBrowse.

Feature & annotation queries

Questions about genes, transcripts, exons, variants, and other genomic features visible in the browser.

Example: “What gene is at this location?” or “Show exons for this transcript.”

Coordinate-based questions

Queries using genomic coordinates (e.g., “What features are at chr1:100000–105000?”).

Track-related questions

What a particular track shows, how to interpret it, or how to change its settings.

Variant interpretation help

Basic explanations of SNPs, indels, and structural variants displayed in tracks.

File & data format questions

Help understanding formats like BAM, CRAM, VCF, GFF, BED as they’re used in JBrowse.

Filtering & search assistance

How to find genes/features by name or apply filters within the browser.

In short, it’s designed to answer context-aware questions about the data and features currently loaded in your JBrowse session, as well as general “how do I…” questions about using JBrowse itself.

You said:
can it tell you which hosted tracks might be useful to answer a specific question?
ChatGPT said:

Yes — it can, but with important limitations.

The jbrowse-plugin-assistant-chatbot is able to:

Suggest relevant types of tracks for a given question
For example, if you ask:

“Is this region conserved?” → it might suggest conservation or comparative genomics tracks

“Are there known variants here?” → it might suggest VCF/variant tracks

“Is this gene expressed?” → it might suggest RNA-seq or expression tracks

However, it cannot reliably:

Browse the internet to discover all externally hosted tracks

Automatically search public hubs (like UCSC track hubs) unless they are already configured or exposed through your JBrowse instance

Guarantee that a suggested hosted track actually exists or is accessible in your environment

What it can do well:

Work with tracks already loaded in your JBrowse session

Recommend categories of tracks you should look for

Help you decide which kinds of hosted data would best answer your biological question

So the practical answer is:

It can recommend the type of hosted tracks you should use, and sometimes point you to already-configured ones — but it doesn’t automatically discover new external hosted tracks.

You said:
can it tell you things like "which are the most significant;y upregulated on this gtrack?"
ChatGPT said:

Short answer: No — not in a fully reliable, “statistical” sense.

The jbrowse-plugin-assistant-chatbot cannot truly compute differential expression or statistical significance from a track on its own. It doesn’t run bioinformatics analysis pipelines. Instead, it can:

What it can do

Help you interpret what you’re seeing on expression tracks (e.g., RNA-seq coverage, bigWig signal tracks).

Help you identify visually strong signals, for example:

“Which genes in this view show much higher signal compared to nearby regions?”

Summarize metadata already present in feature tracks (e.g., fold-change values embedded in GFF/VCF attributes, if they exist).

Filter or sort features that already include statistics, such as:

log2FoldChange

pvalue

padj

What it cannot do

It cannot:

Perform differential expression calculations

Run statistical tests

Compare multiple samples quantitatively unless those values are already encoded in the track’s feature attributes

The realistic answer

If you already have a track that contains differential expression results (like a BED/GFF with log2FC and adjusted p-values), the chatbot can help you find things like:

“Which features have the highest log2FC in this region?”

But if you’re asking it to discover “most significantly upregulated genes” directly from raw coverage tracks (like BAM/bigWig), it can’t do that statistically.

In one sentence:
It can report and filter existing significance data, but it can’t generate significance itself.

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions