Models of codon substitution are developed that incorporate physicochemical properties of amino acids. When amino acid sites are inferred to be under positive selection, these models suggest the nature and extent of the physicochemical properties selected for. This is accomplished by first partitioning the codons based on some property of the amino acids they code for, and then using this partition to parametrize the rates of property-conserving and property-altering base substitutions at the codon level by means of finite mixtures of Markov models that also account for codon and transition:transversion biases. Here, we apply this method to two positively-selected receptors involved in ligand-recognition; the class-I alleles of the human Major Histocompatibility Complex (MHC) of known structure and the S-locus Receptor Kinase (SRK) of the sporophytic self-incompatibility system (SSI) in cruciferous plants (Brassicaceae), whose structure is unknown. Through likelihood ratio tests we demonstrate that the positively selected MHC and SRK proteins are under physico-

Using genomic data from homologous microsatellite loci of pure AC repeats in humans and chimpanzees, several models of microsatellite evolution are tested and compared using likelihood-ratio tests and the Akaike information criterion. A proportional-rate, linear-biased, one-phase model emerges as the best model. A focal length toward which the mutational and/or substitutional process is linearly biased is a crucial feature of microsatellite evolution. We find that two-phase models do not lead to a significantly better fit than their one-phase counterparts. The performance of models based on the fit of their stationary distributions to the empirical distribution of microsatellite lengths in the human genome is consistent with that based on the human-chimp comparison. Microsatellites interrupted by even a single point mutation exhibit a twofold decrease in their mutation rate when compared to pure AC repeats. In general, models that allow chimps to have a larger per-repeat unit slippage rate and/or a shorter focal length compared to humans give a better fit to the human-chimp data as well as the human genomic data.

Abstract-In this study, nonlinear autoregressive exogenous (NARX) models of a heavy-duty single-shaft gas turbine (GT) are developed and validated. The GT is a power plant gas turbine (General Electric PG 9351FA) located in Italy. The data used for model development are three time series data sets of two different maneuvers taken experimentally during the start-up procedure. The resulting NARX models are applied to three other experimental data sets and comparisons are made among four significant outputs of the models and the corresponding measured data. The results show that NARX models are capable of satisfactory prediction of the GT behavior and can capture system dynamics during start-up operation.

Many summary statistics currently used in population genetics and in phylogenetics depend only on a rather coarse resolution of the underlying tree (the number of extant lineages, for example). Hence, for computational purposes, working directly on these resolutions appears to be much more efficient. However, this approach seems to have been overlooked in the past. In this paper, we describe six different resolutions of the Kingman-Tajima coalescent together with the corresponding Markov chains, which are essential for inference methods. Two of the resolutions are the well-known n-coalescent and the lineage death process due to Kingman. Two other resolutions were mentioned by Kingman and Tajima, but never explicitly formalized. Another two resolutions are novel, and complete the picture of a multi-resolution coalescent. For all of them, we provide the forward and backward transition probabilities, the probability of visiting a given state as well as the probability of a given realization of the full Markov chain. We also provide a description of the state-space that highlights the computational gain obtained by working with lower-resolution objects. Finally, we give several examples of summary statistics that depend on a coarser resolution of Kingman's coalescent, on which simulations are usually based.

Abstract-During recent decades, artificial intelligence has been employed as a powerful tool for identification of complex industrial systems with nonlinear dynamics, such as gas turbines (GT). In this study, a methodology based on artificial neural network (ANN) techniques was developed for offline system identification of a low-power gas turbine. The processed data was obtained from a SIMULINK model of a gas turbine in MATLAB environment. A comprehensive computer program code was generated and run in MATLAB environment for creating and training different ANN models with two-layer feed-forward multi-layer perceptron (MLP) structure. The code consisted of various training functions, different number of neurons as well as a variety of transfer (activation) functions for hidden and output layers of the network. It was shown that the optimal model for a two-layer network with MLP structure, consisted of 20 neurons in its hidden layer and used trainlm as its training function, as well as tansig and logsid as its transfer functions for the hidden and output layers. It was also observed that trainlm has a superior performance in terms of minimum MSE, compared with each of the other training functions. The resulting model could predict performance of the system with high accuracy. The methodology provides a comprehensive view of the performance of over 18720 ANN models for system identification of single shaft GT. One can use the optimal ANN model from this study when training from real data obtained from this type of GT. This is particularly useful when real data is only available over a limited operational range.