Autonomous model building with a preponderance of well annotated assay protocols
2 likes622 views
The document discusses advancements in autonomous model building and assay protocols in collaborative drug discovery, particularly focusing on drug efficacy testing for GSK3beta activity. It outlines a methodology for executing high-throughput screening assays, utilizing machine learning and web UI for data curation and validation, while analyzing the results from a significant number of curated assays. Future work aims at enhancing user experience, intensive data mining, and exploring advanced modeling techniques.
Autonomous model building with a preponderance of well annotated assay protocols
- 1. Autonomous model building with a preponderance of well annotated assay protocols Alex M. Clark http://www.bioassayexpress.com ⚡ COLLABORATIVE DRUG DISCOVERY
- 2. Background ◈ Assay protocols are text ◈ Institutions suffering from data overload: ▷ much interest in using consistent markup ▷ semantic web ontologies: BAO, DTO, CLO, GO... ◈ Since last talk: many refinements, some new structure-activity analysis features 2 Cell-Free Homogeneous Primary HTS to Identify Inhibitors of GSK3beta Activity (1) Dispense 1 uL/well of CABPE, 0.5 uL of ATP, and 1 uL of positive control GW8510 or AB in respective wells according to plate design to 1536-well assay ready plates (Aurora 29847) that contain 2.5 nL/well of 10 mM compound using BioRAPTR (Beckman) to start the reaction. Incubate at room temperature for 60 minutes. (2) Add 2.5 uL/well of ADP-glo (Promega, V9103) with BioRAPTR, incubate at room temperature for 40 minutes (3) Add 5 uL/well of ADP-glo (Promega, V9103) with Combi nL (Thermo), incubate at room temperature for 30 minutes
- 3. Curation ◈ Web UI ◈ Machine learning ◈ Common Assay Template ◈ PubChem data 3
- 4. Public Data ◈ Designed Common Assay Template for summary ◈ Data from PubChem: ▷ 3500 assays selected ▷ mostly MLPCN ◈ Curated as a validation set ◈ PubChem also has molecules... 4
- 5. Analysis ◈ Selected: ▷ Secondary assays ▷ Infectious diseases ▷ IC50 results 5
- 6. Analysis ◈ Selected: ▷ Secondary assays ▷ Infectious diseases ▷ IC50 results 5
- 7. Molecules ◈ Per assay: convenient way to view & download... 6
- 20. Grids ◈ View known & calculated measurements, plotting assays vs. compounds... 11
- 21. Grids ◈ View known & calculated measurements, plotting assays vs. compounds... 11
- 22. Real Data Only 12 ◈ Assays are singletons by default, can be grouped...
- 25. Filling in Blanks 13 ◈ Bayesian model created for each column ◈ SAR for all assays used ◈ Missing cells replaced with predictions ◈ Calculations realtime: ▷ model on server ▷ predictions on client
- 26. Trend Detection 14 ◈ Model compound found using the Selectivity option:
- 27. Trend Detection 14 ◈ Model compound found using the Selectivity option: multidrug resistant tuberculosis
- 28. Trend Detection 14 ◈ Model compound found using the Selectivity option: multidrug resistant tuberculosis trypanosomiasis (ROC 0.56)
- 31. Future Work ◈ Current focus is UI/UX and proof-of-concept... ▷ intensive data mining is next: annotations + SAR ▷ data is ready to reveal some amazing discoveries ◈ Crowd curation may be necessary, and more detailed templates ◈ More modelling techniques, far beyond Bayesians: parallel project is underway... 16
- 32. Deployment ◈ Public ▷ http://www.bioassayexpress.com ▷ http://beta.bioassayexpress.com (curation) ◈ Vault ▷ sync with CDD Vault through API ◈ Private ▷ custom integration, installed behind firewall 17
- 33. Acknowledgments ◈ Collaborative Drug Discovery ⇨ Barry Bunin ⇨ Janice Kranz ⇨ Hande Küçük ⇨ Peter Gedeck 18 ◈ And we're hiring ⇨ resumes to: ◈ More information http://github.com/cdd/bioassay-template http://www.bioassayexpress.com http://collaborativedrug.com alex@collaborativedrug.com