Improving genomics-based predictions for precision medicine through active elicitation of expert knowledge

摘要

Predicting the efficacy of a drug for a given individual, usinghigh-dimensional genomic measurements, is at the core of precision medicine.However, identifying features on which to base the predictions remains achallenge, especially when the sample size is small. Incorporating expertknowledge offers a promising alternative to improve a prediction model, butcollecting such knowledge is laborious to the expert if the number of candidatefeatures is very large. We introduce a probabilistic model that can incorporateexpert feedback about the impact of genomic measurements on the sensitivity ofa cancer cell for a given drug. We also present two methods to intelligentlycollect this feedback from the expert, using experimental design andmulti-armed bandit models. In a multiple myeloma blood cancer data set (n=51),expert knowledge decreased the prediction error by 8%. Furthermore, theintelligent approaches can be used to reduce the workload of feedbackcollection to less than 30% on average compared to a naive approach.