1. Benchmarking Methods for Identifying Causal Mutations Tal Friedman

2. Rare Genetic Diseases Our goal: identify and diagnose rare genetic diseases Difficult for clinicians due to incredibly low exposure Often not already documented

3. Overview Why are we doing any of this? Background What you did Why this was important Finding the cause of rare genetic diseases (introduce rare genetic diseases) We have this PC thing, trying to help clinicians find additional patients and candidate genes It uses the HPO to do some fancy stuff But, we don’t really know how well it’s working Exomiser is a good start (how well is it working) 1) try to recreate results, 2) extend to patient matching domain

4. PhenomeCentral Clinicians upload patient data

5. PhenomeCentral Matchmaking algorithm displays most similar patients Get additional evidence from other clinicians

6. Background Phenotype: Observable characteristics Human Phenotype Ontology (HPO) Robinson et. al

7. Exomiser (Robinson et. al, 2014)

8. Objectives Reproduce Exomiser performance Expand to new patient similarity domain

9. Patient Simulation Control Genome Mutation HPO Terms Infected Patient Disease

10. Results

11. Patient Similarity Phenotypic similarity algorithm Hypothesis: same disease/causal gene Combine Exomiser results

12. Patient Pair Simulation Control Genome A Sampled mutation Sampled HPO terms Patient 1 Control Genome B Sampled mutation Sampled HPO terms Patient 2 Disease Phenotypic Noise & Imprecision

13. Results (preliminary)

14. Challenges Data More data

15. Challenges ROC Curve for Phenotypic Similarity Algorithm

16. ROC Curves For a binary classifier Plots TPR vs. FPR for varying threshold values Often compared with AUC

17. Acknowledgements

18. Questions!