Notes Nick Lynch (Unlicensed)
Andrew Conkie, Carlos Faerman , Mark Roberts, Frederik Broek ,Chris Morris, Jens Hoefkens Peter Henstock, Kostas Vougas, Terry Stouch, Ted Slater, Nick Lynch
Support the awareness and adoption of AI within Life Science
What could help groups get started with their strategy and looking ahead
Avoiding pitfalls
Model use best practice (examples good & bad)
What could help those further ahead with adoption
How can we utilise learnings from other industries
To select best practice goals (possible exemplar business use case(s)) and create a business case(s) for funding by mid Feb 2018 to present at London Conference
| Item | Summary |
|---|---|
| Use Cases | Failed Clinical Trials Persuasive models, QSPR Use Case (Chris): druglikeness. With larger datasets, it will be possible to provide a better molecule quality indicator than RO5 or QED. Public data is available from the point of USAN assignment, but not at the hit to lead stage when triage is most important. With suitable methods for ensuring confidentiality of individual molecules, it will be possible to collaborate on a shared model. |
| Implementation | Support those implementing models and a model environment. |
| Reproducibility | Tie the pitfalls to larger issues of Model and Implementation reproducibility |
| Bad AI award. | Make an effort to raise the standard of publication and peer review, by identifying non-reproducible and overfitted results in papers. |
| Training and training material | Should we develop a wider training course? Fundamental part of Life Science capability Who to partner with here? Online Courses (Moots) or add to others (EdX etc) |
| Partners/collaborators | Toolkit providers Platform providers Community sites Service providers |
| Be clear about AI definition | Do we also mean Augmented AI NLP, Machine learning and Neural Networks |
| Asking Pharma on their AI strategies for training/best practice | How best to understand current challenges - see some of the survey outputs in community site |
Examples of pitfalls approaches: Mark R, Andrew C
Short compilation of pitfalls (and especially if there has been an impact or potential impact to data output or decision making)
Aim to show the business value a project that would address these issues