Dx29: assisting genetic disease diagnosis with physician-focused AI pipelines
Download as PPTX, PDF1 like677 views
dx29 is an AI-assisted platform developed by Foundation29 to support the diagnosis of rare genetic diseases, addressing the lengthy average diagnosis time of 5 years. The platform utilizes next-generation genetic sequencing data and natural language processing to assist physicians in navigating complex diagnostic pathways. Currently in beta, dx29 is designed to be a free, cloud-based resource that fosters collaboration and innovation in the diagnosis of rare diseases.
Dx29: assisting genetic disease diagnosis with physician-focused AI pipelines
- 1. Dx29: genetic rare disease diagnosis support with AI Pablo Botas Head of Science, Foundation29
- 2. O R G A N I Z A T I O N P L A T I N U M S P O N S O R S Thank you! C O L L A B O R A T O R S
- 3. [email protected] Rare genetic diseases are very challenging to diagnose, with the average child waiting for diagnosis for 5 years. Next generation genetic sequencing data may hold the key to diagnosis, however analysis can become a paramount task with multiple factors affecting conclusions. Dx29, an AI-assisted platform facilitates this task, allowing the physician to drive the analysis. Dx29 is a free platform developed by Foudation29, in close collaboration with academic groups Pablo Botas Head of Science at Foundation29
- 4. https://globalgenes.org/rare-facts/ 1 in 10 People affected by Rare Disease 3 of 10 Children with a Rare Disease won’t live to see their 5th birthday 8 in 10 Have a genetic origin 4.8 Years For patients to receive a diagnosis 95% of Rare Diseases lack approved treatment 10000 Rare Diseases Rare disease facts 4
- 5. 5
- 6. Lack of diagnosis is lack of data • Lack of diagnosis is about lack of signal • Doctors can’t let you know where you are • Your journey starts by collecting information 6
- 8. Diagnosis is about finding a route • Navigation requires a map and data coordinates • Inaccurate navigation has very high costs!! 8 Required coordinates: 1. Symptoms 2. Genetic Variants
- 9. Icons by different authors from www.flaticon.com Symptom-based path-finding (diagnosis) Rare diseases have complex phenotypes Differential diagnosis based on subtle details Rare diseases: inexperience Disease evolution is gradual Symptom identification is hard Short time per patient, independently of complexity 9
- 10. 10 Licensed under CC BY-SA-NC
- 11. Icons by different authors from www.flaticon.com Genotype-based path-finding (diagnosis) Genetic testing is a black box Mutation analysis is often manual Symptom identification is hard! EMR is not a good communication tool Difficult consideration of new symptoms 11
- 12. : AI as the GPS powering the navigation 12
- 13. New patient to study The navigator DNA data Symptom extraction using NLP 1 Variant prioritization supported by the phenotype 2 Candidate symptoms based on the mutations 3 Candidate conditions for assessment 4 13
- 14. 1. Symptom identification Dx29 uses NCR1,2, developed by the Centre for Computational Medicine at SickKids, Toronto Deep NN to classify concepts into Human Phenotype Ontology (HPO)3 terms EMR, electronic medical records; NN, neural network; NLP, natural language processing 1. Github. 2019. Available from: https://github.com/ccmbioinfo/NeuralCR (Accessed 11 June 2019); 2. Arbabi A, et al. 2019; 7(2):e12596. doi: 10.2196/12596. 3. Köhler S, et al. Nucleic Acids Res 2018;47:D1018–D1027 14
- 15. How to compute on concepts?
- 16. Rank mutation pathogenicity based on public databases: Exomiser1,2 1. Robinson PN, et al. Genome Res 2014;24:340–34; 2. Github. 2019. Available from: https://github.com/exomiser/Exomiser (Accessed 11 June 2019); 3. Zhou X, et al. Nat Commun 2014;5:4212; 4. Wang Genomics Lab. 2019. Available from: http://phenolyzer.wglab.org/ (Accessed 11 June 2019); 5. Yang H, et al. Nat Methods 2015;12:841–843 2. Variant analysis
- 17. 1. Robinson PN, et al. Genome Res 2014;24:340–34; 2. Github. 2019. Available from: https://github.com/exomiser/Exomiser (Accessed 11 June 2019); 3. Zhou X, et al. Nat Commun 2014;5:4212; 4. Wang Genomics Lab. 2019. Available from: http://phenolyzer.wglab.org/ (Accessed 11 June 2019); 5. Yang H, et al. Nat Methods 2015;12:841–843 What if no genetic data is available? Build list of candidate genes based on disease networks3 : Phenolyzer4,5 2. Variant analysis
- 18. 3. Engage with patient: What is missing? 18
- 20. • Open platform to deploy and run diagnosis models • Data available to academia to develop and test new approaches • Open-source code and philosophy • Data donor concept implementation • Diagnosis of complex cases based on clustering using ML Overall platform summary 20
- 21. • Very useful in areas with no access to genetic testing • Cloud-based, no installation required • Available (beta) at dx29.ai • A single variant prioritisation algorithm doesn’t work for all conditions • NLP is an ongoing challenge • Phenotype quality in public databases is poor • Seamless UX is required for a complex process • Patients need their own navigator • Very good feedback from physicians • Exploring symptoms is useful and innovative • High performance: academia
- 22. This is great, but what about the data used … ? 22
- 23. The need for high quality data
- 25. • PSR means the object is a pulsar. • The J reveals that a coordinate system known as J2000 is used. • 1302 and 6350 are coordinates like the latitude and longitude. PSR J1302-6350
- 26. Dravet Syndrome Lennox-Gastaut Syndrome Allan–Herndon–Dudley Syndrome SCN1A MECP2 CDKL5 STXBP1 26 Rett Syndrome
- 28. 28
- 29. SCN1A Dravet Syndrome Febrile Seizures Temporal Lobe Epilepsy Myoclonic Astatic Epilepsy Lennox-Gastaut Syndrome Migrating Partial Epilepsy of Infancy (MMPSI) Autism http://epilepsygenetics.net/the-epilepsiome/scn1a-this-is-what-you-need-to-know/ 29
- 30. 30
- 32. Let’s do a better job!! https://deepai.org/publication/embedding-complexity-in-the-data-representation-instead-of-in-the-model-a-case-study-using-heterogeneous-medical-data
- 33. How to compute on concepts?
- 35. Data locker of patient-owned data, for the development of new algorithms. Patients with complex conditions (multiple genes, big genomic changes) need an approach based on mixed learning : navigator for patients 35
- 36. 36 Final notes 1. Key technologies: NGS + AI 2. NGS+AI makes healthcare personalized, accurate and affordable 3. NGS+AI require development of standards and platforms 4. Seamless UX is vital for technology adoption 5. Transformation of healthcare into information-centered process 6. Rare diseases offer the perfect niche for testing: want to help? 7. Foundation29 is laying the basis for such transformation
- 37. Thanks and … See you soon! Thanks also to the sponsors. Without whom this would not have been posible. O R G A N I Z A T I O N P L A T I N U M S P O N S O R S C O L L A B O R A T O R S
Editor's Notes
- #20: Esta es nueva, quizás demasiado nivel de detalle