Discovery of Logic-Probabilistic Rules from COVID-19 Vaccine Antibody Response in Older People: Results from the GeroCovid VAX Study

This paper presents a novel approach to handle uncertain, noisy and incomplete biomedical data through Probabilistic Logic Programming. Unlike previous systems, based on Inductive Logic Programming (ILP) for rule discovery or on user-defined probabilities, our approach enables the automatic discovery of probabilistic integrity constraints directly from biomedical data. These constraints are annotated with a probability and can assign a probability of belonging to the positive class to logical interpretations, which are used to formalize patients’ medical records. We apply the PASCAL (“ProbAbiliStic inductive ConstrAint Logic”) algorithm for learning such constraints, and we show that it outperforms traditional ILP and data mining approaches, such as TILDE, Aleph and Association Rules, on the data relative to long-term care facility residents participating in the GeroCovid Vax study.