Trialanalysis with Kabuki 1 & 2 snydrom

[aslgraefe]

@Chenjing0525 could you please send me the to do or the resumé of our chat last week how I should proceed?

[Chenjing0525]

You can use the Correlation Workflow and Synergy Workflow to analyze any cohort you're interested in. Feel free to check them out—if anything's unclear, just let me know and I’ll be happy to add more details.

[Chenjing0525]

You can clone my repo to your local repo. And use pip install -e . to install the package. And I already add you to the ppkt2synergy_case_studies repo. You can find the workflow there. The workflow will guide you to use the package.

[aslgraefe]

all right, thank you. I think we agreed on analyising the Kabuki 1 & 2 cohorts of the Phenopacket store, correct @pnrobinson?

[aslgraefe]

@Chenjing0525
I ran the analysis on both kabuki 1 & 2 cohorts from the phenopacket store. you can find the notebook here on the 8-kabuki-1-2-case-study branch: https://github.com/P2GX/ppkt2synergy_case_studies/blob/8-kabuki-1-2-case-study/Notebooks/kabuki_trial/KMT2D_KDM6A.ipynb and the according files in that folder https://github.com/P2GX/ppkt2synergy_case_studies/tree/8-kabuki-1-2-case-study/Notebooks/kabuki_trial.

The analysis for the comprehensive correlation against all cohorts in the phenopacket store threw an error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[10], line 3
      1 correlation_type = "spearman"
      2 # Initialize the HPO statistics analyzer with the filtered HPO matrix and target matrix
----> 3 hpo_analyzer = HPOStatisticsAnalyzer(
      4     (all_hpo_matrix_filtered,target_hpo_matrix[1]), 
      5     min_individuals_for_correlation_test=30)
      7 # Compute the correlation and p-value matrices using Spearman correlation
      8 hpo_analyzer.compute_correlation_matrix(correlation_type, n_jobs=32)

File ~/Documents/git/ppkt2synergy_case_studies/venv/lib/python3.13/site-packages/ppkt2synergy/analysis/hpo_correlation_analyzer.py:82, in HPOStatisticsAnalyzer.__init__(self, hpo_data, min_individuals_for_correlation_test)
     78     raise ValueError("mask must be a pd.DataFrame or np.ndarray")
     80 if relationship_mask.shape[0] != relationship_mask.shape[1] or \
     81     relationship_mask.shape[0] != hpo_matrix.shape[1]:
---> 82     raise ValueError("mask must have the same number of rows and columns as hpo_matrix has features")
     83 if not np.all(np.isin(self.relationship_mask[~np.isnan(self.relationship_mask)], [0])):
     84     raise ValueError("relationship_mask must contain only 0 or NaN")

ValueError: mask must have the same number of rows and columns as hpo_matrix has features

Do you know why?

Regarding the clinical interpretation, I am not sure whether the results are very interesting for this cohort, what do you think? @pnrobinson

[Chenjing0525]

Thanks for running the analysis on both Kabuki 1 & 2 cohorts — the notebook and files on the 8-kabuki-1-2-case-study branch look great.

Regarding the error during the correlation analysis across all cohorts:
I looked into it, and it seems the issue arises because the HPO term IDs and labels may not consistently map one-to-one across the full phenopacket store. This causes mismatches in matrix alignment, especially when use_label=True is used during matrix generation.

I'll work on resolving this by ensuring a consistent and stable ID-to-label mapping across all cohorts.
In the meantime, please use use_label=False when generating the HPO matrices — this will ensure consistency and prevent shape mismatches in correlation analysis.

Let me know if anything else comes up!

[aslgraefe]

Hey, I ran the analysis again - now it worked with comparing against entire Phenopacket store.
@pnrobinson could you please check the results pleasefor clinical significance and/or whether I should run this analysis on a different cohort and/or whether I should write about this in the upcoming manuscript?
here is everything: https://github.com/P2GX/ppkt2synergy_case_studies/tree/8-kabuki-1-2-case-study/Notebooks/kabuki_trial

Also happy to meet this week and talk about it if you like! I could for example come in to the BIH on Thursday for the team meeting.
Best,
Adam

[Chenjing0525]

@aslgraefe Hi Adam,

thanks a lot for running the analysis and sharing the results!
Just to let you know — I’ve made some changes to the package: there is now a new version available which supports labels properly, so you can continue to explore new cohorts with it. I also updated the workflow accordingly, so the new workflow can be used directly. In addition, the plots now have tooltips that allow you to see the data distribution interactively, which should make the exploration more convenient.

Looking forward to seeing what you find!
Best，
Jing

[aslgraefe]

Hi Jing,
great to hear, I'll give it a go a few more times.
Is a manuscript already in preparation? It would help to read more about the tool to understand its functionalities. In there I would be happy to help with writing, too.
Best,
Adam

[Chenjing0525]

Hi Adam,

thanks a lot! The manuscript is still in preparation — I’m currently working on finalizing it. I’ll definitely keep you posted once there’s a draft to share, and it would be wonderful to have your input on the writing.

Best,
Jing