1 of 27

Reproducing Aspergillus

fumigatus Analysis in BRC-Analytics

Variant Calling and Azole Resistance Mutations

Scott Cain

Research Professor

Penn State

scott@scottcain.net

@scottcain@genomic.social

2 of 27

Introduction - Goals for this talk

Introduce BRC-Analytics and Galaxy
Introduce using AI assistants with Galaxy
Focus: Variant calling and identifying mutations related to azole resistance in Aspergillus fumigatus
Scope: Analysis of two BioProjects (22 sequencing runs) with variant calling and resistance-specific SNP identification
Emphasize, that despite access to full compute cluster, these tools are FREE to use!

3 of 27

Platforms Overview

Portal for analyzing organism data��Easy access to ENA/SRA datasets��Seamless integration with Galaxy

Reproducible workflow platform��Backed by TACC for compute��JupyterLite for interactive analysis

4 of 27

Analysis Workflow Steps

Go to brc-analytics.org; Pick organism, assembly, analysis workflow and SRA data; send to Galaxy
“Blindly” accept the parameters sent over from BRC-Analytics, start workflow.
After waiting a few hours to do 200+ computational steps, do a few ad hoc computation steps to identify tandem repeats
Use Gemini’s CLI to generate a JupyterLite notebook to generate GFF of SNPs and TRs
Use Gemini to generate a JupyterLite NB to find resistant-specific SNPs (step towards genotyping)

5 of 27

1. Click 'Organisms'

6 of 27

2. Click 'Filter'

7 of 27

3. Select Aspergillus fumigatus

8 of 27

4. Click 'Analyze'

9 of 27

5. Click 'Variant calling'

10 of 27

6. Click 'Continue'

11 of 27

7. Click 'Enter Accession(s)'

12 of 27

8. Click 'Continue'

13 of 27

9. Click 'select all' checkbox

14 of 27

10. Click 'Add 22 Sequencing Runs'

15 of 27

11. Click 'Launch In Galaxy'

16 of 27

Step 2: Haploid Variant Caller in Galaxy

BRC-Analytics passes info about SRA data, annotation GTF and genome assembly to use for analysis.

Hit “Run Workflow” and Galaxy fetches the data it needs and runs the analysis.

17 of 27

Step 3: Extract & Process cyp51A Region

Ad hoc processing pipeline: BAM Slicing → Assembly → Tandem Repeat Detection

18 of 27

Step 4: AI-Assisted Analysis - cyp51A Variants

AI-Assisted Analysis: Gemini-Generated JupyterLite Notebook

Iterative development with AI assistance��Extracts variants from SnpEff (VCF) files and tandem repeat data��Tracks variants sources back to SRA accessions��Generates GFF3 output

�Enables genome browser visualization

Reproducibility: the JL Notebook is the artifact that does the analysis

(ie, we don’t have to trust that an LLM would do the same thing again)

Side note:

why Gemini?

19 of 27

Step 4: AI-Assisted Analysis - cyp51A Variants

Iterative back and forth with Gemini to produce a JupyterLite notebook that will run in Galaxy to extract variants (SnpEff and etandem repeats) around cyp51A

20 of 27

Step 4: AI-Assisted Analysis - cyp51A Variants

Result: Confirmation of Simmons. The 3 resistant samples all have the A→T/L98H SNP and a tandem repeat in the promoter

21 of 27

Step 5: AI-Generated Set Operations

Step 5: Finding Resistant-Specific Mutations

Susceptible

Union of all SNPs�across susceptible�samples

Resistant

Intersection of�SNPs across resistant�samples

Difference

Resistant SNPs minus�Susceptible SNPs =�Resistance-Specific

Start with the per sample VCFs generated by the variant calling workflow

22 of 27

Step 5: AI-Generated Set Operations

Step 5: Finding Resistant-Specific Mutations

Similar iterative back and forth with Gemini to create a JupyterLite notebook…

23 of 27

Step 5: AI-Generated Set Operations

Step 5: Finding Resistant-Specific Mutations

Output: 765 Resistant Specific SNPs (VCF) - Mutations unique to resistant strains

24 of 27

Conclusions

Successfully reproduced computational analysis from peer-reviewed paper

Leveraged BRC-Analytics + Galaxy for reproducible variant calling workflow

AI assistance (Gemini) accelerated analysis script development

Identified cyp51A mutations associated with azole resistance

… and these tools are all FREE!

25 of 27

Future Directions

Work ongoing to integrate LLMs directly into both BRC-Analytics and Galaxy via MCP servers.

Working on an ‘end to end’ mode: submit jobs from BRC site, autonomously executes in Galaxy, get a report back in BRC-Analytics��Relatedly, working on improved sample sheet handling, primarily to work with differential expression workflows

… and these tools will all still be FREE!

26 of 27

Acknowledgements

The Galaxy Team, including

Dannon Baker

Danielle Callan

Marius van den Beek

John Davis

Michelle Savage

Mike Schatz

Sergei Pond

Anton Nekrutenko

Collaborators at Clever Canary:

Dave Rogers

Mim Hastie

Fran McDade

Hunter Craft

Plus collaborators at UCSC Genome Browser and the Texas Advanced Computing Center

Funding for BRC-Analytis from NIAID and general Galaxy funding from NIH and NSF.

27 of 27

Resources

BRC-Analytics.org

UseGalaxy.org (or .eu or .org.au or .fr etc)

Galaxy history:�https://usegalaxy.org/u/scottcain/h/a-fumigatus-variants

GitHub repo with files used to do this work:�https://github.com/scottcain/brc-galaxy-ai

BRC-Analytics matrix channel:�https://matrix.to/#/#brc-analytics:matrix.org

GitHub repo