To answer these questions we identified the

To answer these questions, we identified the Km and Vmax values for each child from clinical data, based on a pharmacometric approach. The mean population estimates of 0.012μM for Km for isoniazid. In biochemical assays for measurement of Km that examined isoniazid acetylation by human NAT2 from 2-h post-mortem liver homogenates of a rapid acetylator in the laboratory, the Km values were 0.018±0.004μM (Weber and Cohen, 1968). This means that the pharmacometric approach of using the rate of production of N-acetylisoniazid and disappearance of isoniazid is a robust enough approach to calculate these values for each child, and gives values similar to those from enzyme kinetic experiments of liver biopsy specimens. In other words, the measurement of the phenotype was likely accurate. In the case of the NAT2 kinase inhibitors we identified in the children, genotypes such as NAT2∗∗∗∗, and NAT2∗ alleles, and the predominance of the NAT2*5 allele have been shown to be common in Black South African adults (Dandara et al., 2003; Loktionov et al., 2002). Thus, the distribution of the common alleles in our small pediatric cohort mirrors that in adults. Application of AI methods to these parameter estimates and genotypes identified several important patterns between the predictors and enzyme reaction kinetic constants. Our approach is tractable and follows steps that are easy to apply to any pediatric drug whose metabolite is known for each child, and for a population of children. Thus, our approach could be used to examine the effects of maturation and drug doses on both compartmental pharmacokinetic parameters and Michaelis-Menten relationships for other phase II metabolism reactions, an unexplored pediatric space (Kearns et al., 2003). Second, a current problem in systems pharmacology is the difficulty in scaling mathematical models from the level of chemical reactions inside the cell to the level of the individual patient, and to populations of patients. The MARS equations output could allow this to be accomplished, starting with reaction kinetics at the level of a Michaelis-Menten reaction to whole body physiological parameters in the child (isoniazid elimination rate), based on relevant clinical and demographic predictors automatically selected by the algorithm. Moreover, MARS handles both linear and non-linear interactions simultaneously, allowing for construction of non-linear models. Furthermore, MARS is assumption and distribution free and is flexible enough to identify interactions without a prior falsifiable hypothesis to link reactions constants to a predictor, as is the case of isoniazid concentration and Km. The findings led to new hypotheses being generated, which can now be tested in separate studies using standard statistical approaches. Thus, the mathematical relationships are identified first, and then physiological meanings can be investigated further to improve the precision of the estimates for better reproducibility. Third, we found that the relationships between predictors such as age, dose or isoniazid concentration, and NAT2 enzyme kinetics were non-linear. In other words, several clinical and physiological factors affected speed of isoniazid acetylation and affinity, and interacted with each other. Of major importance was the effect of age on both speed of acetylation and enzyme affinity. Age played a critical role in modulating these enzyme activities, and age\'s contribution changed with child\'s age, with the maximal adult type of activity encountered after 5.3years old. This time period likely reflects the time to maturation of the NAT2 enzyme. On the other hand, the effect of dose and drug concentration always superseded that of NAT2 genotype. The effect of dose and drug concentration was persistent regardless of the enzyme reaction constant examined, and was often the primary predictor, except in the case of Km. The physiological and pharmacological basis for this remain to be worked out, but the fact that the hinge was always encountered beyond a certain concentration suggests a real concentration-dependent effect, consistent with many pharmacological processes. The GCV step identified R2≥80% on post-test sets, which means there is a high likelihood of the same observations with future data-sets, so that the findings are likely valid. However, clearly nothing in our current understanding of Km and Vmax can explain how the pharmacological and clinical factors we identified as predictors could affect both enzyme affinity and maximum velocity for isoniazid acetylation. However, the historical disaster of gray baby syndrome in children treated with the drug chloramphenicol suggests that maturation has a major effect on enzyme function (Glazer et al., 1980; Mulhall et al., 1983; Sutherland, 1959). The functional deficiency in the UDP-glucuronyl transferase enzyme system encountered in the first few weeks of life, especially in premature neonates, led to cardiovascular collapse when high concentrations of the antibiotic were achieved in the children (Glazer et al., 1980; Mulhall et al., 1983; Sutherland, 1959). However, we reiterate that, the philosophy behind machine learning methods such as MARS is prediction and pattern recognition, and not identification of causal pathways. Thus, the next step is to test the effect of age and isoniazid concentration on NAT2 Km and Vmax using standard hypothesis testing approaches, and if confirmed then a new understanding of NAT2 catalyzed reaction kinetics should be sought.