Using artificial intelligence and high-throughput hormone measurements
Using artificial intelligence and high-throughput hormone measurements to predict chemical effects on steroidogenesis
US Army Engineer Research and Development Center
We have constructed a Bayesian network to predict which enzymes in the steroidogenesis pathway may be impacted by chemical exposure. Briefly, steroid hormones and enzymes were encoded as nodes while edges between nodes represented conditional probabilities. For instance, two nodes lead into progesterone such that the probability of progesterone production is conditional on probabilities for both pregnenolone and HSD3B1 activity being present. Using this Bayesian network, we can query the likelihood of any represented enzyme in the steroidogenesis pathway being active or inactive. Data for hormone levels were retrieved from the ToxCast high-throughput H295R assay for input. Based on these data where 936 chemicals altered the levels of ≥1 hormone and a subset of 227 chemicals altered the levels of ≥4 hormones, our model predicted specific enzyme inhibition for 178 chemicals, identifying some novel targets for these chemicals and possible mechanisms underlying steroidogenesis disruption.