Discussion
Suggestions for Discussion
- General validation techniques have to be integrated within the framework:
- To be able to perform the required task, learning & prediction has to be kept completely separate from validation techniques
- I.e. the validation approaches are managing the prediction methods
- For this an internal reporting/prediction format has to be found or agreed on
- What kind of format (-> WP1)
- What existing validation routines from project members are out there?
- IST lazar, IDEA AMBIT, NTUA Y-scrambling
- And how to integrate them
- To be able to perform the required task, learning & prediction has to be kept completely separate from validation techniques
- A substantial amount of statistical tests will have to be performed
- Can we agree for R, as underlying statistical toolkit
- Are there other competitive, OpenSource, projects or programs available?
- Greatest challenge: Validation against confidential data
- How can this be achieved?
- Convince advisory board members
- Public and artificial data has to be transferred on a non-networked supply chain
- Test are run locally (in house), some type of reports will have to sent back. however:
- However: Reporting should not allow disclosure of underlying confidential DB
- Is this an issue ?
- How to compare different in-house DBs?
- Is it possible or even wanted, to have a confidential - confidential validation?
- Companies will not share their data
- How can this be achieved?
- Guidelines
- OECD guidelines (http://ecb.jrc.ec.europa.eu/DOCUMENTS/QSAR/OECD_ENV_JM_MONO(2007)2.pdf)
- Which one to incorporate and how?
- Or use ECB QSAR model reporting format (QPRF: http://ecb.jrc.it/qsar/qsar-tools/qrf/QPRF_version_1.1.pdf)
- Is it sufficient?
- OECD guidelines (http://ecb.jrc.ec.europa.eu/DOCUMENTS/QSAR/OECD_ENV_JM_MONO(2007)2.pdf)