Wednesday, March 6, 2024

AI Corner: Can AI Read the FOCR Poster on HRD Harmonization?

Header: FOCR summarized interim results of its HRD harmonization workstream as a poster.  I found the poster hard to get my head around.  Can AI read the poster better than me?

###

For a couple years, Friends of Cancer Research FOCR has been running a  major project on HRD harmonization  - homologous recombination deficiency testing.   Today I got a long email from them reporting results, and, linking to a PDF "poster presentation" to summarize results from two experiments.  The first was an "in silico" study, the second was a "patient sample" study.   

See the FOCR HRD home page here:

https://friendsofcancerresearch.org/hrd/#ProjectOutcomes

I found the poster challenging:


However, one take home lesson is on the right, that whether one used percent aggreement or percent disagreement, these were in the 70-80% range.  You can rate agreement data this way; it's hard to say "right or wrong" as there isn't a gold standard.

Here are three AI readings of the poster:

CHAT GPT 4

The scientific poster presents findings from the Friends of Cancer Research HRD Harmonization Project, which analyzed 13 assays for measuring Homologous Recombination Deficiency (HRD) in 90 high-grade serous ovarian tumors. Key findings include:

  • The median HRD positivity rate was consistent with prior publications, 49% for in silico analysis and 53% for patient sample analysis.
  • Inter-assay agreement on HR status was variable, with altered BRCA1/2 samples more uniformly called HRD.
  • Positive/negative agreement varied across assays, with modest to high levels of agreement.
  • The influence of BRCA1 and BRCA2 alterations on HRD calls was significant, indicating the importance of these genes in HRD status determination.

The conclusion emphasizes the importance of understanding assay agreement to inform assay interpretation and improve alignment of HRD scores, facilitating better treatment decisions. Future steps include further analysis on the impact of various factors on HRD call concordance and hosting a public meeting to discuss findings and recommendations.

###

CLAUDE.AI

Here are the key findings and take-home lessons from the scientific poster on the HRD (Homologous Recombination Deficiency) harmonization project:

Key Findings:

  • There was significant variability in HRD positivity rates across different assays, ranging from 23-74% (median 53%) for patient samples and 9-67% (median 49%) for in silico samples.
  • Agreement on HRD status calls between assay pairs was variable, with modest to high levels of positive and negative percent agreement.
  • Samples with BRCA1/2 mutations were more uniformly called HRD positive across assays compared to wild-type BRCA1/2 samples.
  • Inclusion of different factors (e.g., genomic scars like LOH, mutational signatures, non-BRCA1/2 gene mutations) in assay algorithms did not strongly explain the variability in HRD calls for in silico samples.

Take-Home Lessons:

Significant inter-assay variability exists in determining HRD status, which can impact treatment decisions for drugs like PARP inhibitors.

  • BRCA1/2 mutation status strongly influences HRD calls, with higher agreement for mutation-positive samples.
  • Understanding the sources of variability (e.g., algorithms, biomarkers used) is crucial for improving alignment of HRD scores and assay interpretation.
  • Further research is needed to examine the impact of clinical factors, sample quality, and specific gene alterations on HRD call concordance.
  • Harmonization efforts and recommendations from this project will inform future use of HRD assays to improve treatment selection for patients.

In summary, while HRD testing can guide targeted therapy, there are significant inter-assay differences that need to be addressed through harmonization and understanding the key drivers of variability for optimal clinical implementation.

###
PERPLEXITY

Key Findings from the HRD Harmonization Study:

  • Study Objective: The research analyzed 13 independently performed assays to measure Homologous Recombination Deficiency (HRD) using 90 high-grade serous ovarian tumors.
  • Participants: The study involved 16 organizations representing 18 HRD assays, including commercial and academic developers.
  • Methodology: Assay developers analyzed TCGA ovarian cancer samples in silico and patient samples to measure HR status and contributing factors for each sample.

Results:

  1. In Silico Sample Results: Variability in HR status determination was observed across assays, with correlations among continuous HR scores varying substantially. The median HRD positivity rate was consistent with prior publications.
  2. Patient Sample Results: Inter-assay agreement on HR status calls was variable, influenced by the inclusion of factors related to causes versus consequences in HRD score determination. Samples with altered BRCA1/2 showed higher Positive Predictive Agreement (PPA) and lower Negative Predictive Agreement (NPA).

Conclusions:

  • Understanding the agreement among assays is crucial for interpreting results accurately and making informed treatment decisions.
  • Future research should focus on examining the impact of clinical and sample factors on HRD call concordance and provide recommendations for the future use of HRD assays.

Take-Home Lessons:

  • Interpretation Challenges: Variability in HR status determination across assays highlights the need for standardization to improve clinical interpretation.
  • BRCA1/2 Influence: Samples with altered BRCA1/2 genes demonstrated higher agreement levels, emphasizing the impact of these genes on HRD calls.
  • Clinical Implications: Enhancing assay alignment and understanding factors influencing HRD scores can aid in guiding appropriate treatment decisions for patients.

###

What do I think?

Only CLAUDE noted that gross percent positive ranged from around 10-20% for one test to 60-70% for the highest test.  To me, that sounds like an important (if frightening) fact.  Imagine if one doctor sends samples to a lab that calls 10% positive (1 in 10), and another doctor sends samples to a lab that calls 70% positive (7 in 10).

However, those numbers define the highest highest and lowest extremes, over the range of tests, the average agreement was better circa 75%.   

CHAT GPT and PERPLEXITY (but not Claude) noted that 13 tests (labs) were tested, which is key helpful fact for me to understand what was done.

###

Definitions and Explanations of HRD

What if you don't know HRD from a hole in the ground?  Separate from the poster, I asked Perplexity to define HRD, show the testing used, and explain why a doctor needs it. Note, in the response, it footnoted 5 sources.  It should be open access:

https://www.perplexity.ai/search/Provide-a-definition-t3xFrB8bQzGZaEHdWyYQwA

###  

Perplexity Explains HRD to Us:

Definition of HRD (Homologous Recombination Deficiency) and Testing Methods

HRD, or Homologous Recombination Deficiency, is a cellular phenotype characterized by the inability of a cell to effectively repair DNA double-strand breaks using the homologous recombination repair (HRR) pathway. This deficiency can lead to genomic instability and contribute to cancerous tumor growth 

Molecular Steps or Methods Used to Test for HRD:

  1. Genomic Scars Analysis: HRD testing involves assessing genomic scars that indirectly measure HRD, reflecting permanent genomic changes caused by DNA repair deficiency. This includes:
    1. Loss of heterozygosity (gLOH): Permanent loss of one parent's allele copy of a gene at a specific part.
    2. Number of telomeric imbalances (TAI): Regions in a chromosome with allelic imbalance.
    3. Large-scale transitions (LST): Genomic alterations involving chromosome breakages
  2. Comprehensive Genomic Profiling (CGP): CGP using Next-Generation Sequencing (NGS) can assess causal genes and genomic scarring together in one assay. 
    1. It helps identify rare mutations and other relevant genomic signatures in various cancers like ovarian, breast, prostate, and pancreatic cancers

PARP Inhibitor Testing: HRD testing can guide treatment decisions, especially for PARP inhibitors. If a tumor is positive for HRD, it may respond well to PARP inhibitors, which block the enzyme's action in cancer cells, preventing them from repairing their damaged DNA and causing cell death

Importance of HRD Testing for Cancer Patients:

  1. Treatment Personalization: HRD testing provides crucial information for personalizing treatment options for cancer patients, guiding the selection of therapies like PARP inhibitors or platinum-based chemotherapy based on the HRD status of the tumor
  2. Prognostic Insights: It offers insights into the disease course and helps identify variants that may cause HRD, aiding in predicting outcomes and planning comprehensive treatment strategies
  3. Family Risk Assessment: By detecting germline mutations of BRCA1/2, HRD testing can also provide predispositional insights into family members' risk of developing ovarian cancer and other cancers

In summary, HRD testing plays a vital role in oncology by enabling personalized treatment decisions, offering prognostic insights, and assessing familial cancer risks based on genetic mutations and genomic instability markers.

##

Not AI.  

See HRD at plain old Wikipedia, here:

https://en.wikipedia.org/wiki/Homologous_recombination

It's helpful to see again what "homologous recombination" is, to start with (HR), before factoringin a 'deficiency" of it (HR-D)..

###