One of the most important goals of seismic stratigraphy studies is to interpret the elements of the seismic facies with respect to the geologic environment. Prestack seismic data carry rich information that can help us get higher resolution and more accurate facies maps. Therefore, it is promising to use prestack seismic data for the seismic facies recognition task. However, because each identified object changes from the poststack trace vectors to a prestack trace matrix, effective feature extraction becomes more challenging. We have developed a novel data-driven offset-temporal feature extraction approach using the deep convolutional autoencoder (DCAE). As an unsupervised deep learning method, DCAE learns nonlinear, discriminant, and invariant features from unlabeled data. Then, seismic facies analysis can be accomplished through the use of conventional classification or clustering techniques (e.g., K-means or self-organizing maps). Using a physical model and field prestack seismic surveys, we comprehensively determine the effectiveness of our scheme. Our results indicate that DCAE provides a much higher resolution than the conventional methods and offers the potential to significantly highlight stratigraphic and depositional information.

To characterize the antigenic epitopes of the hemagglutinin (HA) protein of H1N1 influenza virus, a panel consisting of 84 clones of murine monoclonal antibodies (mAbs) were generated using the HA proteins from the 2009 pandemic H1N1 vaccine lysate and the seasonal influenza H1N1(A1) vaccines. Thirty-three (39%) of the 84 mAbs were found to be strain-specific, and 6 (7%) of the 84 mAbs were subtype-specific. Twenty (24%) of the 84 mAbs recognized the common HA epitopes shared by 2009 pandemic H1N1, seasonal A1 (H1N1), and A3 (H3N2) influenza viruses. Twenty-five of the 84 clones recognized the common HA epitopes shared by the 2009 pandemic H1N1, seasonal A1 (H1N1) and A3 (H3N2) human influenza viruses, and H5N1 and H9N2 avian influenza viruses. We found that of the 16 (19%) clones of the 84 mAbs panel that were cross-reactive with human respiratory pathogens, 15 were made using the HA of the seasonal A1 (H1N1) virus and 1 was made using the HA of the 2009 pandemic H1N1 influenza virus. Immunohistochemical analysis of the tissue microarray (TMA) showed that 4 of the 84 mAb clones cross-reacted with human tissue (brain and pancreas). Our results indicated that the influenza virus HA antigenic epitopes not only induce type-, subtype-, and strain-specific monoclonal antibodies against influenza A virus but also cross-reactive monoclonal antibodies against human tissues. Further investigations of these cross-reactive (heterophilic) epitopes may significantly improve our understanding of viral antigenic variation, epidemics, pathophysiologic mechanisms, and adverse effects of influenza vaccines.

Epitopes serve an important role in influenza infection. It may be useful to screen universal influenza virus vaccines, analyzing the epitopes of multiple subtypes of the hemagglutinin (HA) protein. A total of 40 monoclonal antibodies (mAbs) previously obtained from flu virus HA antigens (development and characterization of 40 mAbs generated using H1N1 influenza virus split vaccines were previously published) were used to detect and classify mAbs into distinct flu virus sub-categories using the ELISA method. Following this, the common continuous amino acid sequences were identified by multiple sequence alignment analysis with the GenBank database and DNAMAN software, for use in predicting the epitopes of the HA protein. Synthesized peptides of these common sequences were prepared, and used to verify and determine the predicted linear epitopes through localization and distribution analyses. With these methods, nine HA linear epitopes distributed among different strains of influenza virus were identified, which included three from influenza A, four from 2009 H1N1 and seasonal influenza, and two from H1. The present study showed that considering a combination of the antigen-antibody reaction specificity, variation in the influenza virus HA protein and linear epitopes may present a useful approach for designing effective multi-epitope vaccines. Furthermore, the study aimed to clarify the cause and pathogenic mechanism of influenza virus HA-induced flu, and presents a novel idea for identifying the epitopes of other pathogenic microorganisms.

No abstract

In this paper, a novel type of neural network called a grey radial basis function network (GRBFN), is addressed and applied to reduce the temperature influence on the output of fibre optic gyroscopes (FOGs) and to improve their performance. The reasons why grey theory is introduced into the RBF neural network are based on two facts. First, the output of FOGs is affected greatly by environmental factors, especially by environmental temperature variation, hence there will be large randomness on output data. Second, the modelling performance will be affected by the randomness inherent in the output data of FOGs when a neural network approach is applied to the model. That is, poor performance results from large randomness and vice versa. The grey accumulated generating operation (AGO), a basis of the grey theory, is reported to possess a randomness reduction property. Because of these facts, the GRBFN model is presented and expected to have better modelling precision of temperature drift in FOGs. The proposed GRBFN model consists of the grey AGO, the RBF neural network, and the grey inverse accumulated generating operation (IAGO). Raw data are first preprocessed by the grey AGO and then put into the RBF network to perform modelling. Modelling results of the GRBFN are then obtained from the grey IAGO. The numerical results of real drift data under different temperatures from a certain type of FOG verify the effectiveness of the proposed GRBFN model powerfully. The RBF neural network modelling approach is also investigated to provide a comparison with the GRBFN model. Under identical training conditions, the GRBFN’s training speed has been enhanced greatly. It is shown that the proposed GRBFN model with different network structures outperforms the RBF network itself.