In clinical predictive modeling, recommender-system algorithms (collaborative-filtering-style methods originally developed for consumer recommendation, e.g., market-basket/association-rule approaches such as a priori algorithms) can be adapted to biomedical outcome prediction by treating patients as "users" and genomic or cytogenetic abnormalities as "items," learning combinatorial patterns of co-occurring features associated with an outcome. This approach is presented as an alternative to conventional univariate/multivariate statistical association testing, with the claim that it can identify predictive feature combinations applicable to a larger fraction of a patient cohort while achieving higher classification accuracy than single-marker associations alone.
In clinical predictive modeling, recommender-system algorithms (collaborative-filtering-style methods originally developed for consumer recommendation, e.g., market-basket/association-rule approaches…