MOTIVATION: Gene expression plays a crucial role in cell function, and enhancers can regulate gene expression precisely. Therefore, accurate prediction of enhancers is particularly critical. However, existing prediction methods have low accuracy or rely on fixed multiple epigenetic signals, which may not always be available. RESULTS: We propose a two-stage framework that accurately predicts enhancers by flexibly combining multiple epigenetic signals. In the first stage, we designed a Blending-KAN model, which integrates the results of various base classifiers and employs Kolmogorov-Arnold Networks (KAN) as a meta-classifier to predict enhancers based on flexible combinations of multiple epigenetic signals. In the second stage, we developed a Stacking-Auto model, which extracted sequence features using DNABERT-2 and located the enhancers based on the Stacking strategy and AutoGluon framework. The accuracy of the Blending-KAN model reached 99.69 ± 0.11% when five epigenetic signals were used. In cross-cell line prediction, the accuracy was more significant than or equal to 93.72%. With Gaussian noise, it still maintains an accuracy of 98.74 ± 0.03%. In the second stage, the accuracy of the Stacking-Auto model is 80.50%, which is better than the existing 17 methods. The results show that our models can be flexibly used to predict and locate enhancers utilizing a combination of multiple epigenetic signals. AVAILABILITY AND IMPLEMENTATION: The source code is available at https://github.com/emanlee/Hi-Enhancer and https://doi.org/10.6084/m9.figshare.29262158.v1.

With the development of deep sequencing, recent studies indicate that a miRNA precursor can generate multiple miRNA isoforms (isomiRs). The family prediction of canonical miRNAs and isomiRs could provide a basis for miRNA functional research. In this study, we propose a novel method for family identification of canonical miRNA and isomiRs based on incremental learning. First, a benchmark dataset is constructed by processing data based on miRNA sequences and their family annotation. Moreover, sequence embedding and RNN are used for capturing essential features and inherent dependencies within a sequence. Finally, incremental learning is applied to accommodate the continuous influx of miRNA sequencing data, enabling RNN to stay relevant and effective over time. Comparative experiments and ablation studies illustrate the effectiveness of our model, which can help to comprehensively understand miRNA’s function.

The Convolutional Neural Network (CNN) model excels at learning local features, but struggles with capturing global large-scale features, particularly in extremely heterogeneous areas. In contrast, the Graph Convolution Network (GCN) is an effective tool for PolSAR image classification, demonstrating a capability to learn large-scale global features proficiently.To learn effective features for both heterogenous terrain objects and edge details well, a novel region partition based hybrid deep network is proposed for adaptive learning features for boundary and non-boundary regions, which can learn both large-scale global features for extremely heterogeneous terrain objects and pixel-wise features for edge details. The proposed method can effectively partition a PolSAR image into boundary and non-boundary regions, and design a CNN and GCN subnetworks for them respectively. Subsequently, a unified network is designed to effectively fuse both the advantages of GCN and CNN to enhance classification performance. The experiments verify the proposed algorithm can achieve better performance than compared methods in both region homogeneity and boundary preservation.

Human activity recognition (HAR) based on wearable sensors is a hot topic in health detection and motion management. Nonetheless, conventional identification methods necessitate substantial labeled datasets, and the acquisition of high-quality labeled data crucial for human activity recognition is both time-consuming and costly. To tackle this problem, transfer learning is used to annotating unlabeled or a few labeled target domains using labeled source domains. Meanwhile, individuals exhibit significant differences in amplitude, angles, and other aspects when performing the same action. Due to the limited expressive capacity of time series, effectively capturing various types of differences simultaneously poses a challenge, impacting the effectiveness of transfer. Therefore, in this paper, we propose a stacked transfer network (STN) for 2D modeling and feature extraction of sensor data in steps. It adaptively decomposes complex temporal variations into multiple intra- and inter-periodic variations using the Fast Fourier Transform (FFT), allowing the model to focus more on the trends of the activty and less on the effects of individual discrepancy. Our comprehensive cross-domain activity recognition (CDAR) experiments on three large public activity recognition datasets (i.e., OPPORTUNITY, PAMAP2, and UCI-DSADS) show that the STN achieves high accuracy in activity recognition ttransfer.

Queue counting using WiFi channel state information (CSI) faces challenges due to susceptibility to external factors and relies on ideal testing environments for current methods. We propose an efficient CSI recurrence plot (RP)-based framework for queue counting (CRPF-QC), containing a transformation module and a recognition module. The conversion module transforms the CSI into RP, distinct from traditional models using a single signal point as the unit for feature extraction, utilizing the signal changes at different timestamps as units for feature extraction and effectively preserving the amplitude and phase relationships between any two time points. In the recognition module, the convolutional neural network (CNN) and the long short-term memory (LSTM) network are combined to profoundly understand the internal structure and changes within the image. The proposed integration framework is adept in the automatic extraction of amplitude and phase features, therefore improving image recognition accuracy. Meanwhile, we explore dynamic changes in the queuing crowd detection based on the Fresnel zone theory, identifying individuals’ entering and exiting behaviors at different positions within the Fresnel zone and updating the count accordingly, which makes up for the shortcomings of the static model. Intensive evaluations demonstrate that CRPF-QC, employing just two layers of CNN and one layer of LSTM, excels in adapting to dynamic environmental changes, outperforming traditional queue counting methods. Additionally, the dynamic model attains a perfect 100% accuracy in both scenarios.

As the largest class of small non-coding RNAs, piRNAs primarily present in the reproductive cells of mammals, which influence post-transcriptional processes of mRNAs in multiple ways. Effective methods for predicting piRNA and mRNA target relationships can help identify piRNA functions, investigate the possibility of piRNAs as biomarkers and therapeutic targets. In this study, we propose a computational approach for classifying the relationships of piRNA-mRNA pairs based on an interactive inference network (IIN). First, we gather piRNA-mRNA target data, collect sequence data by position alignment, and construct a benchmark dataset. Furthermore, a reliable negative set is constructed by positive-unlabeled learning. Finally, we view a piRNA and a mRNA sequence as a premise and hypothesis sentence, respectively, and IIN model is used to predict the relationship between them. The experiments demonstrate that our method effectively characterizes piRNA-mRNA interaction and could be beneficial for researchers to investigate piRNA functions.

When clustering gene expression, it is expected that correlation coefficients of genes in the same clusters are high, and that gene ontology (GO) enrichment analysis of most clusters will be significant. However, existing short-term gene expression clustering algorithms have limitations. To address this problem, we proposed a novel clustering process based on angular features for short-term gene expression. Our method (named AngClust) uses angular features to indicate the change of trend in gene expression levels at two neighboring time points. The changes of angles at multiple time points reflects the change of trend of the overall expression levels. Such changes are used to measure whether the expression trends of different genes are similar. To obtain functionally significant clusters from the clustering results, we evaluated numbers of genes in clusters, average correlation coefficient, fluctuation, and their correlation with GO term enrichment. The efficacy of AngClust outperform two other measures, Euclidean distance (ED) and dynamic time warping of correlation (DTW), on a dataset of yeast gene expression. The ratios of GO and pathway term-enriched of clusters of AngClust is higher than or equal to that of STEM and TMixClust on human, mouse, and yeast time series of gene expression.

As a key equipment to switch the direction of a running train, railway turnout works in complex condition which makes its fault diagnosis difficult. Generally, existing methods identify the fault by analyzing the turnout action curve acquired by sensors, which have certain practical value for fault diagnosis, but poor practicability for varied types like double or multiple action turnout. In this paper, fault detection is carried out according to the distance between the normal current curve and the test curve calculated by fast dynamic time warping algorithm. In view of the singular point problem involved, a segmentation method for current curve based on the key nodes in the turnout conversion process is proposed and applied to the fault detection of single action and double action turnouts. Experimental results show that proposed approach can effectively improve the matching accuracy of adaptive diagnosis model which is more than 96%. Furthermore, compared with the traditional dynamic time warping algorithm, the time cost can be reduced by more than 5 times.

The samples representing abnormal situation is usually very few in the dataset, which makes it difficult to learn the features of abnormal samples by machine-learning-based methods. To improve the accuracy of anomaly detection, the number of abnormal samples should be expanded to ensure the balance of the dataset. In this paper, a discrete synthetic minority oversampling technique (D-SMOTE) is proposed to generate new samples. A closed area is constructed using the three nearest abnormal samples in the dataset. The new samples are then uniformly interpolated in a closed area. By this means, the problem of the imbalance for the original dataset is handled, thus improving the data quality. Based on the expanded datasets, a two-dimensional convolutional neural network (2D CNN) is constructed to detect abnormal samples. In experiments, three cases and different machine learning methods are considered for comparison. Several indexes including accuracy, precision, confusion matrix, F1-score, and Recall have been used to evaluate the detection effectiveness. The results show that the abnormal samples can be detected accurately using oversampling data obtained from the proposed D-SMOTE method.

PIWI-interacting RNAs (piRNAs) are a type of small non-coding RNAs which bind with the PIWI proteins to exert biological effects in various regulatory mechanisms. A growing amount of evidence reveals that exosomal piRNAs are potential biomarkers for diagnosis and treatment of complex diseases. Effective methods for the prediction of exosomal piRNAs are the foundation of piRNA functional research. In this study, we propose an end-to-end deep network for identifying exosomal piRNAs based on features learned from natural language processing (NLP) models for sequence embedding with attention mechanism. First, a benchmark dataset is constructed by processing piRNA subcellular localization annotated data and sequence data. Moreover, bagging positive unlabeled learning is applied to get the reliable negative set. Finally, we treat a piRNA sequence as a sentence and its k-mer subsequence as a token. Sequence embedding models with self-attention mechanism is designed to extract features from exosome piRNA sequences, which are used for the prediction task. Compared with three competing methods, our model achieves the best performance and reveals the key factors of exosomal piRNA sequences by the attention mechanism. Our model characterizes exosomal piRNAs and could be beneficial for researchers to investigate exosomal piRNAs’ functions.

Conventional decision systems, when applied to policy selection in autonomous driving, can introduce unsafe factors due to uncertainty. Thus, in this paper, we introduce and implement an improved algorithm Advantage Actor-Critic Method of Actor-Critic Method to improve the safety of autonomous driving policy selection, and give a specific push-to process, including the combination of advantage function with Reinforce with Baseline algorithm, we applied the algorithm to the agent of the scenario designed in this paper, and militarily expanded the design based on the autopilot scenario, the agent will be trained by the A2C policy-value network learning the rules of the designed scenario for adversarial gaming, and the value of the loss function is reduced from about 4.5 to about 1.7; and also with the game tree related algorithm for gaming in the pentad scenario, the win rate are above 60% Finally, the algorithm has better performance in strategy selection from the game results.

Fault detection of point machine operations is discussed in this paper, which is critical for ensuring the safety of a running train.This paper adopts the method of data enhancement and convolutional neural network (CNN) to study and realize the fault diagnosis of power data of switch switch machine. Firstly, six typical fault types and possible fault causes are summarized by analyzing the working process of switch machine and its power curve characteristics. In view of the imbalance of switch data, the synthesized minority oversampling technique (SMOTE) is implemented to generate switch fault data and balance switch data set. In view of the low accuracy of turnout fault diagnosis, one-dimensional convolutional neural network is adopted to classify the turnout fault diagnosis model, which further improves the accuracy of turnout fault diagnosis model and provides theoretical support for railway field maintenance. To a certain extent, it overcomes the difficulties of instability and low efficiency of manual turnout fault detection method.

At a time when end-to-end speech recognition technology is becoming more and more popular, we conduct research on various end-to-end speech technologies, and use the Transformer-based speech framework to study and find that its multi-head attention is not effective in local feature acquisition. And in the face of noise problems in real scenes, the training convergence speed is too slow. In order to solve the problems caused by Transformer, a new speech recognition framework based on MCNN-Transformer-CTC speech recognition method is proposed. Through MCNN (multi-stream convolutional neural network) in the pre-acoustic unit through multiple parallel channels Local feature extraction is carried out in terms of time width and spectral capability, which makes up for the lack of self-attention mechanism in local feature extraction, and the multitask learning method is used to add CTC structure to make up for the problem of slow training convergence. The training effect of this model on the Aishell1 dataset has reached a CER of 6.23%, which is a further improvement compared to the Transformer model.

Compared to conventional cameras, the new type of vision camera-neuromorphic cameras, which can avoid motion blur and have the advantages of high spatiotemporal resolution, high dynamic range, low latency, etc. In this paper, two motion compensation methods are introduced for operating event flow based on differential neuromorphic camera, one is motion compensation based on event counting image and time image, another is motion compensation based on image of warped events. Comprehensive experimental study is carried on two datasets, the motion compensation method based on event counting image and time image is better at generating clear motion compensation images, highlighting moving objects inconsistent with the background motion, while the motion compensation method based on image of warped events tends to generate motion compensated images of moving objects with more prominent or clearer edges. Simultaneously, two motion compensation methods are further applied to realize the object detection method based on threshold or maximum contrast. The former method can be adapted in many cases, while the latter method can well detect the translational moving target.

Abstract feature dimension and cross-domain classification in text classification may lead to low efficiency of text classification. Therefore, in order to study the specificity of the 2019-novel coronavirus, this paper proposes an improved clustering structure based on deep sparse auto-encoder for cross-domain text classification. In this structure, the word vector model and cosine similarity are used to construct the similarity matrix, and then the deep sparse automatic encoder based on unsupervised learning is used to reduce the dimension and extract the feature structure of complex network, and the k-means clustering method is used for testing. Finally, the results are obtained through mean-shift autonomous classification. The performance of the structure is verified on the data set of the title of the paper. The experimental results show the effectiveness of the structure and the practicability of the paper.

Cleft palate (CP) is the second most common congenital birth defect. The etiology of CP is complicated, with involvement of various genetic and environmental factors. To investigate the gene regulatory mechanisms, we designed a powerful regulatory analytical approach to identify the conserved regulatory networks in humans and mice, from which we identified critical microRNAs (miRNAs), target genes and regulatory motifs (miRNA-TF-gene) related to CP. Using our manually curated genes and miRNAs with evidence in CP in humans and mice, we constructed miRNA and transcription factor (TF) co-regulation networks for both humans and mice. A consensus regulatory loop (miR17/miR20a-FOXE1-PDGFRA) and eight miRNAs (miR-140, miR-17, miR-18a, miR-19a, miR-19b, miR-20a, miR-451a and miR-92a) were discovered in both humans and mice. The role of miR-140, which had the strongest association with CP, was investigated in both human and mouse palate cells. The overexpression of miR-140-5p, but not miR-140-3p, significantly inhibited cell proliferation. We further examined whether miR-140 overexpression could suppress the expression of its predicted target genes (BMP2, FGF9, PAX9 and PDGFRA). Our results indicated that miR-140-5p overexpression suppressed the expression of BMP2 and FGF9 in cultured human palate cells and Fgf9 and Pdgfra in cultured mouse palate cells. In summary, our conserved miRNA-TF-gene regulatory network approach is effective in detecting consensus miRNAs, motifs, and regulatory mechanisms in human and mouse CP.

Optimal approximation of linear system models is an important task in the controller design and simulation for complex dynamic systems. In this paper, we put forward a novel nature-based meta-heuristic method, called artificial raindrop algorithm, which is inspired from the phenomenon of natural rainfall, and apply it for optimal approximation of a stable linear system. It mimics the changing process of a raindrop, including the generation of raindrop, the descent of raindrop, the collision of raindrop, the flowing of raindrop and the updating of raindrop. Five corresponding operators are designed in the algorithm. Numerical experiment is carried on the optimal approximation of a typical stable linear system in two fixed search intervals. The result demonstrates better performance of the proposed algorithm comparing with that of other five state-of-the-art optimization algorithms.