We screened 26 bisphosphonates against a farnesyl diphosphate synthase from Plasmodium vivax, finding a poor correlation between enzyme and cell growth inhibition (R(2) = 0.06). To better predict cell activity data, we then used a combinatorial descriptor search in which pIC(50)(cell) = a pIC(50)(enzyme) + bB + cC + d, where B and C are descriptors (such as SlogP), and a-d are coefficients. R(2) increased from 0.01 to 0.74 (for a leave-two-out test set of 26 predictions). The method was then further validated using data for nine other systems, including bacterial, viral, and mammalian cell systems. On average, experimental/predicted cell pIC(50) correlations increased from R(2) = 0.28 (for an enzyme-only test set) to 0.70 (for enzyme plus two descriptor test set predictions), while predictions based on scrambled cell activity had no predictive value (R(2) = 0.13). These results are of interest since they represent a general way to predict cell from enzyme inhibition data, with in three cases, R(2) values increasing from approximately 0.02 to 0.72.