Neural network model successfully predicts amphetamine effects in new task

We have previously shown that neural network models can simulate neuromodulatory effects of catecholamines and account for known D-amphetamine effects on performance in a signal detection task. In such models, D-amphetamine effects are simulated as an increase in signal-to-noise ratio (gain) of neuron-like units in regions responsible for selective attention. We modeled human performance in a speeded response competition task with speed/accuracy trade-off never before tested with D-amphetamine (the Eriksen task). The model predicted specific changes in the reaction-time (RT) distribution and time-accuracy curve following D-amphetamine. We evaluated this prediction in a study of eight healthy volunteers using a pre-post placebo-controlled cross-over design. Predicted RT distributions and time-accuracy curves at different levels of gain in the selective attention module were fitted to the empirical data obtained from each subject before and after drug and placebo. This procedure provided a best-fit gain value (Maximum Likelihood Estimate) for each run of each subject. An ANOVA on the best-fit gain values then confirmed the prediction: drug-placebo × pre-post F (1,7) = 8.39 (p<0.01). The significant interaction confirmed that D-amphetamine effects on reaction time distributions and time-accuracy curves were reliably captured by an increase in signal-to-noise ratio of neural processing in brain areas responsible for selective attention. This study thus illustrates how neural network modeling of individual subject data can be tested statistically to evaluate a mechanistic account of drug effects on cognition.