Fig. 2: Optimizing antibiotic selections with linear programming.

Patient feature vectors are ingested by personalized antibiogram models (a) to produce antibiotic efficacy estimates (b). Each patient in the test set receives a predicted probability of efficacy for each antibiotic. In this illustration, pentagons refer to one antibiotic option and triangles refer to another. Green indicates the antibiotic option is likely to cover the patient, orange indicates the antibiotic is unlikely to cover the patient. A linear programming objective function is specified with a set of constraints that limit how frequently certain antibiotics can be used. Here the objective function specifies to maximize the total predicted antibiotic efficacy (green) across the two patients subject to the constraint that each antibiotic option is only used once. c Depicts all possible antibiotic allocations color coded by patient specific antibiotic efficacy estimates produced by personalized antibiograms. Antibiotics allocations are only considered (d) if they meet the constraints of the linear programming formulation. The antibiotic allocation that maximizes the total predicted efficacy across the set of patients (e) is chosen.