A talk presented at NIH-STRIDES on the Future of Biomedical Knowledge Extraction, Data Security, and Data Structures

1. MIT Licensed
Customized by Mountain Genomics in collaboration
with the bioinformatics community.
Version 1.0
Addressing Privacy Concerns
in the Age of Federated Data
Access

2. MIT Licensed
Customized for Mountain Genomics in collaboration
with the bioinformatics community.
Version 1.0
Sections
Part 1: Federated Computing and Identity
Part 2: Data Indexing
Part 3: Data Structures
Part 4: Extracting Knowledge

3. Disclosures
The idea and material presented here is not the view of NCBI, NLM, NIH or any other federal agency. Some of
the underlying work was supported by the Intramural Research Program of the NLM and various other federal
agencies.
Ben Busby is also an advisor to:
● Johns Hopkins
● Ariel Precision Medicine
● Deloitte
through Mountain Genomics, a company headquartered in Pittsburgh, PA.
Ben Busby will work primarily for DNANexus starting February 3rd, 2020; with limited other engagements

4. Think of Data
Security as a
Dynamic Process,
not a Wall

5. Academia:

6. Make it worth it for
investigators and
stakeholders to identify
themselves

7. Making Compliance and Authentication Worthwhile:
Four Illustrative Topic Areas
1
Indexing Data for Federated Discovery on
any Platform, Anywhere in the World
Flexible presentation (API) of viral protein
domains, host-pathogen interactions and
eventually graph loops as proof-of-principle
data federation.
2
Setting up Federated Compute for Secure
and Reproducible Science
Unified control and accounting of compute
from a single node
3
Genome Graphs
Generation of simple, usable systems to not
compare an individual patient or organism to
another -- or small group of -- individual, but
to an entire community, dramatically
compress data, and immediately find “toxic
paths”.
4
Extracting Knowledge from Data
Annotation of haplotypes, instead of graphs,
to allowing investigators to query complex
disease

8. If you build
it, they will
won’t come

9. A collage of
reproducibility
solutions

10. A collage of
infrastructure

11. Not to mention
platforms

12. There is no king of
the mountain!

13. Minimizing attack
surfaces

14. Don’t Make Unnecessary Copies of Data!

15. Satisfying
regulatory
requirements

17. Computing
Across Infrastructure

18. Through
reproducibility
solutions

19. And platforms...

20. Making it worth it for
investigators and
stakeholders to identify
themselves

24. Indexing Data for Federated Discovery on
any Platform, Anywhere in the World!

25. Indexing Data for Federated Discovery on
any Platform, Anywhere in the World!

26. Indexing Data for Federated Discovery on
any Platform, anywhere in the World!
Christiam Camacho, Sej
Modha, Alex Efremov,
Joan Marti-Carreras

27. Indexing Data for Federated Discovery on
any Platform, anywhere in the World!
Christiam Camacho, Sej
Modha, Alex Efremov,
Joan Marti-Carreras

28. A Brief Illustration of Utility for
Human Data, using
Graph Genomes
as an Example

29. Genome Graphs
Initially leveraging standard workflows for annotation of primary datasets
in cloud infrastructure mapped to linear genomes, we will likely transition to
genome graphs over the next ten years. This will allow us to compress
reads, do phasing automatically, do reference-guided assembly on complex
genomes, and detect “toxic paths” automatically. Most importantly, this will
allow us to see complex genotype - phenotype interactions easily.

30. Evan Biederstedt, Alex Dilthey
many others

31. Erik Garrison, many others

32. Justin Zook, many others

33. Eric Dawson, Fernanda Forterre, many
others

34. https://github.com/NCBI-Codeathons/SWIGG
Jason Chin, Alex Gener, many others

35. Medical Annotations of Graphs and Haplotypes
Lon Phan, Sara Kalla, many others

36. Haplocravat2
Mike Smallegan, Kyle
Moad, Matt
Hynes-Grace, Dina
Mikdadi, many others

37. Vince Carey, John Didion, many others

40. A Brief Proposal for Graph-Based Sequence
Compression and Phasing (personal opinion)
++

41. A Brief Proposal for Graph-Based Sequence
Compression and Phasing (personal opinion)
=
(no longer available on
CD-ROM)
In this case, the sequence
data is not PII without
linked metadata

42. Infrastructure for Extracting
Information from Complex Data
Structures!

43. Archive Pipelines, not
Intermediate Datasets

44. Annotate and Support from Existing
Infrastructure

45. Derive Constituent
Data Sources and
Paths from
Aggregate Data
Sources

46. Represent and
Query
Annotation
Relationships

47. Rank Order
Reliability of
Knowledge

48. Linking Knowledge Graphs
and Genome Graphs

49. Reference
Materials!
It is my sincere hope that no one
named Justin gets too mad at me
for making this slide.

50. Thanks to a lot of fancy people for their work on this!
Rob Edwards
Ahmed Moustafa
Surya Saha
Rob Davey
Jason Williams
Fernanda Forterre
Juerven Bolleman
Nuala O’leary
Nick Cooley
Joyce Lee
Yuriy Skripchenko
Felix Shaw
Adrien Cantu
Sonia Gavasso
Michael Crusoe
Lisa Federer
Bhavya Papudeshi
Jordan Izenga
Alexis Norris
Others cited in the talk!
So Many Hackathoners!
We’ll continue working on it at the
upcoming Baylor SV/Pangenomics
Hackathon and other events!
https://www.hgsc.bcm.edu/events/hacka
thon
https://biohackathons.github.io