From: Toward better benchmarking: challenge-based methods assessment in cancer genomics
Advantages | Limitations | Participation barriers |
---|---|---|
Reduction of over-fitting | Narrower scope compared to traditional open-ended research | Incentives not strong enough to promote participation |
Benchmarking individual methods | Ground truth needed for objective scoring | No funding available to support time spent participating in challenges |
Impartial comparison across methods using same datasets | Mostly limited to computational approaches | Fatigue resulting from many ongoing challenges |
Fostering collaborative work, including code sharing | Requires data producers to share their data before publication | Time assigned by organizers to solve a difficult challenge question may be too short |
Acceleration of research | Sufficient amount of high-quality data needed for meaningful results | Lack of computing capabilities |
Enhancing data access and impact | Large number of participants not always available | New data modality or datasets that are too complex or too big poses entry barrier |
Determination of problem solvability | Challenge questions may not be solvable with data at hand | Challenge questions not interesting or impactful enough |
Tapping the `Wisdom of Crowds’ | Traditional grant mechanisms not adequate to fund challenge efforts | Cumbersome approvals to acquire sensitive datasets |
Objective assessment | Difficulties to distribute datasets with sensitive information | Â |
Standardizes experimental design | Â | Â |