Prediction of response to antiretroviral therapy by human experts and by the EuResist data-driven expert system (the EVE study).

OBJECTIVES: The EuResist expert system is a novel data-driven online system for computing the probability of 8-week success for any given pair of HIV-1 genotype and combination antiretroviral therapy regimen plus optional patient information. The objective of this study was to compare the EuResist system vs. human experts (EVE) for the ability to predict response to treatment. METHODS: The EuResist system was compared with 10 HIV-1 drug resistance experts for the ability to predict 8-week response to 25 treatment cases derived from the EuResist database validation data set. All current and past patient data were made available to simulate clinical practice. The experts were asked to provide a qualitative and quantitative estimate of the probability of treatment success. RESULTS: There were 15 treatment successes and 10 treatment failures. In the classification task, the number of mislabelled cases was six for EuResist and 6-13 for the human experts [mean±standard deviation (SD) 9.1±1.9]. The accuracy of EuResist was higher than the average for the experts (0.76 vs. 0.64, respectively). The quantitative estimates computed by EuResist were significantly correlated (Pearson r=0.695, P<0.0001) with the mean quantitative estimates provided by the experts. However, the agreement among experts was only moderate (for the classification task, inter-rater κ=0.355; for the quantitative estimation, mean±SD coefficient of variation=55.9±22.4%). CONCLUSIONS: With this limited data set, the EuResist engine performed comparably to or better than human experts. The system warrants further investigation as a treatment-decision support tool in clinical practice.

Training a perceptron in a discrete weight space.

Learning in a perceptron having a discrete weight space, where each weight can take 2L+1 different values, is examined analytically and numerically. The learning algorithm is based on the training of the continuous perceptron and prediction following the clipped weights. The learning is described by a new set of order parameters, composed of the overlaps between the teacher and the continuous/clipped students. Different scenarios are examined, among them on-line learning with discrete and continuous transfer functions. The generalization error of the clipped weights decays asymptotically as exp(-Kalpha(2)) in the case of on-line learning with binary activation functions and exp(-e(|lambda|alpha)) in the case of on-line learning with continuous one, where alpha is the number of examples divided by N, the size of the input vector and K is a positive constant. For finite N and L, perfect agreement between the discrete student and the teacher is obtained for alpha~Lsqrt[ln(NL)]. A crossover to the generalization error approximately 1/alpha, characterizing continuous weights with binary output, is obtained for synaptic depth L>O(sqrt[N]).

Multilayer neural networks with extensively many hidden units.

The information processing abilities of a multilayer neural network with a number of hidden units scaling as the input dimension are studied using statistical mechanics methods. The mapping from the input layer to the hidden units is performed by general symmetric Boolean functions, whereas the hidden layer is connected to the output by either discrete or continuous couplings. Introducing an overlap in the space of Boolean functions as order parameter, the storage capacity is found to scale with the logarithm of the number of implementable Boolean functions. The generalization behavior is smooth for continuous couplings and shows a discontinuous transition to perfect generalization for discrete ones.