Neuroscience studies have revealed that the brain encodes visual content and embeds information in neural activity. Recently, deep learning techniques have facilitated attempts to address visual reconstructions by mapping brain activity to image stimuli using generative adversarial networks (GANs). However, none of these studies have considered the semantic meaning of latent code in image space. Omitting semantic information could potentially limit the performance. In this study, we propose a new framework to reconstruct facial images from functional Magnetic Resonance Imaging (fMRI) data. With this framework, the GAN inversion is first applied to train an image encoder to extract latent codes in image space, which are then bridged to fMRI data using linear transformation. Following the attributes identified from fMRI data using an attribute classifier, the direction in which to manipulate attributes is decided and the attribute manipulator adjusts the latent code to improve the consistency between the seen image and the reconstructed image. Our experimental results suggest that the proposed framework accomplishes two goals: (1) reconstructing clear facial images from fMRI data and (2) maintaining the consistency of semantic characteristics.

Neuroscience studies have revealed neural processes involving rhythm perception, suggesting that brain encodes rhythmic sounds and embeds information in neural activity. In this work, we investigate how to extract rhythmic information embedded in the brain responses and to decode the original audio waveforms from the extracted information. A spatiotemporal convolutional neural network is adopted to extract compact rhythm-related representations from the noninvasively measured magnetoencephalographic (MEG) signals evoked by listening to rhythmic sounds. These learned MEG representations are then used to condition an audio generator network for the synthesis of the original rhythmic sounds. In the experiments, we evaluated the proposed method by using the MEG signals recorded from eight participants and demonstrated that the generated rhythms are highly related to those evoking the MEG signals. Interestingly, we found that the auditory-related MEG channels reveal high importance in encoding rhythmic representations, the distribution of these representations relate to the timing of beats, and the behavior performance is consistent with the performance of neural decoding. These results suggest that the proposed method can synthesize rhythms by decoding neural representations from MEG.

In this study, we proposed a new end-to-end convolutional neural network, called MS-SincResNet, for music genre classification. MS-SincResNet appends 1D multi-scale SincNet (MS-SincNet) to 2D ResNet as the first convolutional layer in an attempt to jointly learn 1D kernels and 2D kernels during the training stage. First, an input music signal is divided into a number of fixed-duration (3 seconds in this study) music clips, and the raw waveform of each music clip is fed into 1D MS-SincNet filter learning module to obtain three-channel 2D representations. The learned representations carry rich timbral, harmonic, and percussive characteristics comparing with spectrograms, harmonic spectrograms, percussive spectrograms and Mel-spectrograms. ResNet is then used to extract discriminative embeddings from these 2D representations. The spatial pyramid pooling (SPP) module is further used to enhance the feature discriminability, in terms of both time and frequency aspects, to obtain the classification label of each music clip. Finally, the voting strategy is applied to summarize the classification results from all 3-second music clips. In our experimental results, we demonstrate that the proposed MS-SincResNet outperforms the baseline SincNet and many well-known hand-crafted features. Considering individual 2D representation, MS-SincResNet also yields competitive results with the state-of-the-art methods on the GTZAN dataset and the ISMIR2004 dataset. The code is available at https://github.com/PeiChunChang/MS-SincResNet.

Cancer is a fatal disease. It is worth noting that the treatment of cancer still lacks effective drugs for cancer resistance. The development of multi-target drugs is an important direction in the future. Human epidermal growth factor receptor (EGFR and HER2) and Heat shock protein 90 (HSP90) have been proven to be useful targets in various cancer cell lines. To develop dual-target drugs for these two proteins may be more effective in cancer treatment. We performed ligand-based QSAR modeling to select potential TCM candidate compounds for HER2/HSP90 inhibition. The results show that cyclokoreanine B, dehydropodophyllotoxin, alloimperatorine, wanpeinine A, zierin, N-demethylnoracronycine, desacetyleupaserrin, dianthramine, gnoscopine, and formononetin might have the potential for HER2/HSP90 inhibition.

Maize is a well-studied crop. It has been used as a model plant for C4 studies of photosynthesis, as its leaves possess the Kranz Structure (KS). Unfortunately, only few studies addressed the use of computational models to describe dry maize. In particular, the mechanism of KS formation remains unclear during leaf development. This study aims to develop a computational model to answer the following two questions for leaf development in dry maze: (1) How Auxin inhibits BDL, and (2) How the MP transcription activates BDL in the seed of dry maize in early stages of embryonic leaves. We first analyze dry maize based on the S-systems model and compare it with two different regulatory networks: (1) Auxin inhibits BODENLOS (BDL), and (2) MONOPTEROS (MP) activates BODENLOS (BDL). Our hypotheses are: (1) Auxin does not inhibit BDL, and (2) MP does not activate BDL. In the second stage, we compare the S-systems parameter estimation method (SPEM) and the engineering method to analyze the two regulatory networks. Our result suggests a general mechanism for studying how the transient accumulation of Auxin activates self-sustaining and how, similar to other genetic switches, it results in unequivocal developmental responses of leaves in dry maize. The MP activates BDL are very important to the Auxin signaling mediated by MP and BDL proteins which are essential for cell-fate specification events in early embryogenesis of maize.

Investigating the identity, distribution, and evolution of bird species is important for both biodiversity assessment and environmental conservation. The discrete wavelet transform (DWT) has been widely exploited to extract time-frequency features for acoustic signal analysis. Traditional approaches usually compute statistical measures (e.g., maximum, mean, standard deviation) of the DWT coefficients in each subband independently to yield the feature descriptor, without considering the intersubband correlation. A new acoustic descriptor, called the local wavelet acoustic pattern (LWAP), is proposed to characterize the correlation of the DWT coefficients in different subbands for birdsong recognition. First, we divide a variable-length birdsong segment into a number of fixed-duration texture windows. For each texture window, several LWAP descriptors are extracted. The vector of locally aggregated descriptors (VLAD) is then used to aggregate the set of LWAP descriptors into a single VLAD vector. Finally, principal component analysis (PCA) plus linear discriminant analysis (LDA) are employed to reduce the feature dimensionality for classification purposes. Experiments on two birdsong datasets show that the proposed LWAP descriptor outperforms other local descriptors, including linear predictive coding cepstral coefficients, Mel-frequency cepstral coefficients, perceptual linear prediction cepstral coefficients, chroma features, and prosody features. Furthermore, the proposed LWAP descriptor, followed by VLAD encoding, PCA plus LDA feature extraction, and a simple distance-based classifier, yields promising results that are competitive with those obtained by the state-of-the-art convolutional neural networks.

The construction and understanding of Gene Regulatory Networks (GRNs) are among the hardest tasks faced by systems biology. To infer gene regulatory networks from gene expression data has been a vigorous research area. It aims to constitute an intermediate step from exploratory to gene expression analysis. In recent years, many reverse engineering methods have been proposed. In practice, different model approaches will generate different network structures. Therefore, it is very important for users to assess the performance of these algorithms. We present a comparative study with three different reverse engineering methods, including the S-system Parameter Estimation Method (SPEM), the Graphical Gaussian Model (GGM) and the TimeDelay-ARACNE. Our approach consists of the analysis of real gene expression data with the different methods, and the assessment of algorithmic performances by sensitivity, specificity, precision and F-score.

Epilepsy is a chronic neurological disorder that is characterized by recurrent unprovoked seizures. These seizures are due to abnormal, excessive or synchronous neuronal activity in the brain. For a neural network, such as brain, nonlinearity is necessary to descript the complexity of dynamic system. In this study, we compared some nonlinear dynamic indictions, such as Hurst exponent, sample entropy, and detrended fluctuation in time series of epilepsy electroencephalogram regarding different physiological and pathological brain states. We found that The Hurst exponent did not differ between healthy volunteers and intracranial patients (p≫0.05). The sample entropy value did not differ between healthy volunteers and seizure active patients (p≫0.05). In other cases we found statistical significant differences between investigated data sets. We concluded that using nonlinear dynamic indications we could discriminate the electroencephalogram regarding different physiological and pathological brain states of epilepsy patients.

BACKGROUND: Genome-wide identification of specific oligonucleotides (oligos) is a computationally-intensive task and is a requirement for designing microarray probes, primers, and siRNAs. An artificial neural network (ANN) is a machine learning technique that can effectively process complex and high noise data. Here, ANNs are applied to process the unique subsequence distribution for prediction of specific oligos. RESULTS: We present a novel and efficient algorithm, named the integration of ANN and BLAST (IAB) algorithm, to identify specific oligos. We establish the unique marker database for human and rat gene index databases using the hash table algorithm. We then create the input vectors, via the unique marker database, to train and test the ANN. The trained ANN predicted the specific oligos with high efficiency, and these oligos were subsequently verified by BLAST. To improve the prediction performance, the ANN over-fitting issue was avoided by early stopping with the best observed error and a k-fold validation was also applied. The performance of the IAB algorithm was about 5.2, 7.1, and 6.7 times faster than the BLAST search without ANN for experimental results of 70-mer, 50-mer, and 25-mer specific oligos, respectively. In addition, the results of polymerase chain reactions showed that the primers predicted by the IAB algorithm could specifically amplify the corresponding genes. The IAB algorithm has been integrated into a previously published comprehensive web server to support microarray analysis and genome-wide iterative enrichment analysis, through which users can identify a group of desired genes and then discover the specific oligos of these genes. CONCLUSION: The IAB algorithm has been developed to construct SpecificDB, a web server that provides a specific and valid oligo database of the probe, siRNA, and primer design for the human genome. We also demonstrate the ability of the IAB algorithm to predict specific oligos through polymerase chain reaction experiments. SpecificDB provides comprehensive information and a user-friendly interface.