For those of you who are interested in going forward: In essence, this hackathon is an extreme experiment in Personalized medicine.
o Problem statement.
What's wrong with Bill Paseman's kidney and how can he fix it?
o Resources available
- Bill's DNA, RNA, co-morbidities, dental records(TBD) and familial genetic associations.
- Praxis Platform Data sources (Gtex etc. available on the Praxis Biomedical Datasets learning path)
Note: Everything you need to utilize these resources is in the platform (http://www.prxai.com/platform) including the training to become familiar with the tools and datasets available.
o Hackathon Goals
We wil test either your own hypothesis or ones from the list at the end of this post.
Either way, at the end of the hackathon, you will present a technical report to the patient that presents the hypothesis and shows evidence supporting it (or not supporting it).
Here's a brief history of our hackathon results to date.
- May 2018 hackathon: Sv.ai and rarekidneycancer.org held their first p1RCC hackathon with 17 teams. Each team was given my tumor and normal DNA data. They then presented “genes of interest” 3 days later. Several teams, working independently, recommended the same gene of interest. So “discovery overlap” was used as a scoring function to decide which of the 119 total “genes of interest” to pursue after the event. This Hackathon produced one paper from the Clemson team, published in scientific reports.
- March 2020 hackathon: Sv.ai and rarekidneycancer.org held their second p1RCC hackathon with 2 teams. Each team was given my tumor and normal RNA-seq data several months in advance. One team (Lead by Reed Bender from Clemson’s Alex Feltus Lab) produced a list of 18,000 over and under expressed genes in a normalized cohort containing my tumor and those of 5 TCGA patients ("Clemson's 2020 normalized cohort"). The second team, GeneXpress (lead by Jeannette Koschmann), then used Clemson's 2020 normalized cohort to make therapeutic recommendations. This Hackathon produced one paper from the Clemson team, published in Cell.
- November 2020 followup: QuantumInsights.io combined Clemson's 2020 normalized cohort with a TCGA pan-cancer dataset and discovered that my data clustered close to Thyroid Cancer. This was interesting since unbeknownst to QuantumInsights.io, I had a sibling diagnosed with Thyroid Cancer earlier in the year.
- February 2021 followup: I ranked candidate genes from the 2018 hackathon using Clemson's 2020 normalized cohort. In essence, the 2020 differential expression list functioned as a “holdout set” to score the 2018 teams' results. One 2018 team, Biomarkers.ai, headed by Rutger’s Dr. Saed Sayad stood out. We dicuss the reasons for his success including data (GEO), tools (BIOada.com + Linear Discriminant Analysis) and team size (3 members) in this post. But our key takeaway is that reimagining hackathon teams as ensembles of classifiers lets us apply ensemble learning techniques to monitor, speed up and direct the research process after the hackathon is done.
- March 2021: Rutger’s Dr. Saed Sayad used his BIOada platform to suggest Baicalein and Valproic Acid as therapeutic options.
- March 2021 followup: Eva Comperat discussed several new kidney cancer classifications at GUASCO 2021, including TLFRCC (Thyroid-Like Follicular Renal Cell Carcinoma). Given the November 2020 results, I combined the genes described in Ko’s "Whole-genome and transcriptome profiling of a metastatic thyroid-like follicular renal cell carcinoma" with Clemson's 2020 normalized cohort. However, none of Clemson's 2020 normalized cohort had abnormal expression of TLFRCC biomarkers. So if I have a genetic connection to thyroid cancer, it is likely not via TLFRCC.
- April 2022: After taking a leading role in my last two hackathons, Dr Alex Feltus is holding his own hackathon on April 1, 2022. I will participate and take advantage of a co-morbidity (a meningioma) to personally try and validate the following hypotheses:
- Hypothesis: In some genetic backgrounds, Type I Papillary Renal Cell Carcinoma and meningioma are caused by common genetic lesions, e.g. NF2 or UMOD.
- Hypothesis: Type I Papillary Renal Cell Carcinoma and meningioma are caused by disruption of NF2 function.
- Hypothesis: Type I Papillary Renal Cell Carcinoma and meningioma are caused by disruption of UMOD function. (Note: Bill has a UMOD mutation).
- Hypothesis: Meningioma growth is caused by high Uric Acid Levels (Note: Bill has high Uric Acid Levels).
- Hypothesis: Type I Papillary Renal Cell Carcinoma has common genetic lesions with TLFRCC and/or Thyroid Cancer (See here).
- Hypothesis: Whatever hypothesis you come up with!
Sign up if you are interested and feel free to contact me for more information: firstname.lastname@example.org