Daniel Himmelstein, PhD

Data-Driven Drug Repurposing Workshop
2023-10-03

Integrating All Biomedical Knowledge to Systematically Find the Best Opportunities:

From Hetionet to Related Sciences

slides.com/dhimmel/related-sciences

Workshop Information

Data-Driven Drug Repurposing Workshop: Unlocking disease biology and advancing systematic approaches

• CZI Headquarters
• Hosted by CZI & Every Cure
• Redwood City, CA
• 2023-10-03 at 1:30 PM

Session 2: Curating and Representing Biomedical Knowledge in Network-Based Approaches
Session Abstract: Biomedical knowledge graphs (KGs) represent rich relationships and semantics between drugs, targets and diseases, which support novel methods for inferring biological pathways and predicting drug-target links across a network of interacting genes and proteins. In this session, panelists will share efforts to build and apply KGs for drug repurposing, including advances in optimizing predictions as well as challenges in extracting biomedical knowledge and overcoming inconsistencies, contradictions and other limitations and complexities.

Baranzini Lab @ UCSF

Sandler Neurosciences Center

Sergio

PhD in Biological & Medical Informatics

http://www.greenelab.com/

• knowledge graph for drug repurposing
   
• integrates 29 public resources
  knowledge from millions of studies
   
• ~50 thousand nodes
  11 types (labels)
   
• ~2.25 million relationships
  24 types

Hetionet v1.0

Systematic integration of biomedical knowledge prioritizes drugs for repurposing
Daniel S Himmelstein, Antoine Lizee, Christine Hessler, Leo Brueggeman, Sabrina L Chen, Dexter Hadley, Ari Green, Pouya Khankhanian, Sergio E Baranzini
eLife (2017) https://doi.org/cdfk

Project Rephetio

Systematic integration of biomedical knowledge prioritizes drugs for repurposing
Daniel S Himmelstein, Antoine Lizee, Christine Hessler, Leo Brueggeman, Sabrina L Chen, Dexter Hadley, Ari Green, Pouya Khankhanian, Sergio E Baranzini
eLife (2017) https://doi.org/cdfk

1,538 connected

138 connected

disease modifying treatments
+755, −208,413
AUROC = 97.4%

Project Rephetio: Does bupropion treat nicotine dependence?

• Bupropion was first approved for depression in 1985
   
• In 1997, bupropion was approved for smoking cessation
   
• Can we predict this repurposing from Hetionet? The prediction was:
  • 99.5th percentile for nicotine dependence
  • probability 2.50-fold greater than null

Compound–causes–Side Effect–causes–Compound–treats–Disease

Compound–binds–Gene–associates–Disease

Compound–binds–Gene–participates–Pathway–participates–Disease

connectivity search

Hetnet connectivity search provides rapid insights into how biomedical entities are related
Daniel Himmelstein, Michael Zietz, Vincent Rubinetti, Kyle Kloster, Benjamin Heil, Faisal Alquaddoomi, Dongbo Hu, David Nicholson, Yun Hao, Blair Sullivan, Michael Nagle, Casey Greene

GigaScience (2023) https://doi.org/gsd85n

enriched metapaths

enriched paths

• What we learned?
  • unsupervised is hard
     
• Hetionet wishlist
  • time resolution
  • automated updates
  • greater disease coverage
  • data on real world entities

continuing the journey at

how to build biotechs that fail less often?

RS Facets

1. Time-Resolved Biomedical Atlas
   70+ public and private data sets
2. Feature Design
   100s of metrics designed to quantitatively assess risk and reward for 250 million target-disease pairs
3. Predictive ML Models
   Outcomes like probability of clinical success, commercial interest, or economic value creation
4. Validation and Back-testing
   Enable objective comparative performance benchmarking and iterative improvement

RS Facets™ ingests all activities in global biomedicine to systematically predict the best new drug discovery opportunities.

Identifying 1,000s of the very best opportunities from 250 million

Back-testing performance covering all indications except infectious diseases and oncology

RS Facets back-testing and validation data for its Q1 2023 clinical prediction model build; comparisons reflect the performance the Facets system’s top predictions on a bet- and time-matched basis would have had in a given historical year, based only on what was known in that year.

Likelihood of FDA Approval from Phase I

• diving into one data team project at Related Sciences
• EFO OTAR Slim
• cause of the tangle
  • diagnosable diseases
  • grouping terms
• How many diseases are there?

classifying EFO nodes

predicted EFO classifications

Node outline shows precision

• low = dotted
  disease groupings
• medium = dashed
  diseases
• high = solid
  disease subtypes

efo_otar_slim_v3.57.0_precisions.tsv

nxontology software suite

• nxontology

  NetworkX-based Python library for representing ontologies.

• nxontology-ml
  Machine learning to classify ontology nodes.
• nxontology-data
  Making ontologies accessible as simple JSON files.
  • EFO
  • MeSH
  • HGNC Gene Groups
  • PubChem Classifications

Figure from the obo-community slack by Philip Strömert generated with Midjourney prompt:

We cannot interpret our research data anymore because we did not annotate it with ontologies

feature group importance

poll

Which feature group has the greatest influence on the outcome?

Slido: 1342 945

• topology
• cross-references
• prefixes
• subsets
• descriptions
• gpt tags
• gwas