Search

Background: Determination of genome-wide DNA methylation is significant for both basic research and drug development. As a key epigenetic modification, this biochemical process can modulate gene expression to influence the cell differentiation which can possibly lead to cancer. Due to the involuted biochemical mechanism of DNA methylation, obtaining a precise prediction is a considerably tough challenge. Existing approaches have yielded good predictions, but the methods either need to combine plenty of features and prerequisites or deal with only hypermethylation and hypomethylation., Results: In this paper, we propose a deep learning method for prediction of the genome-wide DNA methylation, in which the Methylation Regression is implemented by Convolutional Neural Networks (MRCNN). Through minimizing the continuous loss function, experiments show that our model is convergent and more precise than the state-of-art method (DeepCpG) according to results of the evaluation. MRCNN also achieves the discovery of de novo motifs by analysis of features from the training process., Conclusions: Genome-wide DNA methylation could be evaluated based on the corresponding local DNA sequences of target CpG loci. With the autonomous learning pattern of deep learning, MRCNN enables accurate predictions of genome-wide DNA methylation status without predefined features and discovers some de novo methylation-related motifs that match known motifs by extracting sequence patterns.

Motivation Single-cell DNA methylation sequencing can assay DNA methylation at single-cell resolution. However, incomplete coverage compromises related downstream analyses, outlining the importance of imputation techniques. With a rising number of cell samples in recent large datasets, scalable and efficient imputation models are critical to addressing the sparsity for genome-wide analyses. Results We proposed a novel graph-based deep learning approach to impute methylation matrices based on locus-aware neighboring subgraphs with locus-aware encoding orienting on one cell type. Merely using the CpGs methylation matrix, the obtained GraphCpG outperforms previous methods on datasets containing more than hundreds of cells and achieves competitive performance on smaller datasets, with subgraphs of predicted sites visualized by retrievable bipartite graphs. Besides better imputation performance with increasing cell number, it significantly reduces computation time and demonstrates improvement in downstream analysis. Availability and implementation The source code is freely available at https://github.com/yuzhong-deng/graphcpg.git. [ABSTRACT FROM AUTHOR]

Spiking neural networks (SNNs) is a promising learning model due to its computational efficiency for discrete spike events. However, because of the binary output of spiking neurons, the standard backpropagation (BP) method is not suitable for deep SNN training. In this paper, we design a gradient-based spiking neuron named Relaxation Leaky Integrate-and-Fire (RLIF) neuron to enable the direct training of deep SNNs. Specifically, we develop a Re-weighted Straight-Through-Estimator (Re-STE) to model the RLIF neuron and SNNs. The Re-STE decomposes the output of the famous LIF neuron and generates a spike by iteratively re-weighting the membrane potential (MP). Then, the RLIF neuron is built, and the re-weighted MP provides a surrogate gradient for BP training. We theoretically prove that the Re-STE is an approximation of the STE technique with logarithmic activation. Such property guarantees that the designed RLIF neuron mitigates the mismatching gradient problem within limits. Finally, we design the RLIF-based VGG-like and ResNet-like SNNs for classification tasks. Experimental results show that the RLIF-based SNNs perform well on the popular benchmark datasets (MNIST, CIFAR10, CIFAR100, N-MNIST, CIFAR10-DVS, and DVS128-GESTURE) and achieve significant efficiencies in most cases. [ABSTRACT FROM AUTHOR]

Spiking neural networks are efficient computation models for low-power environments. Spike-based BP algorithms and ANN-to-SNN (ANN2SNN) conversions are successful techniques for SNN training. Nevertheless, the spike-base BP training is slow and requires large memory costs, while ANN2SNN needs many inference steps to obtain good performance. In this paper, we propose an A ctivation C onsistency C oupled A NN- S NN (A C 2 AS) framework to train the SNN in a fast and memory-efficient way. The A C 2 AS consists of two components: (a) a weight-shared architecture between ANN and SNN and (b) spiking mapping units. Firstly, the architecture trains the weight-shared parameters on the ANN branch, resulting in fast training and low memory costs for SNN. Secondly, the spiking mapping units are designed to ensure that the activation values of the ANN are the spiking features. As a result, the activation consistency is guaranteed, and the classification error of the SNN can be optimized by training the ANN branch. Besides, we design an adaptive threshold adjustment (ATA) algorithm to decrease the firing of noisy spikes. Experiment results show that our A C 2 AS-based models perform well on the benchmark datasets (CIFAR10, CIFAR100, and Tiny-ImageNet). Moreover, the A C 2 AS achieves comparable accuracy under 0. 625 × time steps, 0. 377 × training time, 0. 27 × GPU memory costs, and 0. 33 × spike activities of the Spike-based BP model. The code is available at https://github.com/TJXTT/AC2ASNN. • Proposed the A C 2 AS framework for fast and memory efficient SNN training. • Designed an adaptive threshold adjustment algorithm and two spiking mapping units to ensure that the SNN can be equivalently represented by the ANN under the A C 2 AS framework. • The SNNs trained by A C 2 AS achieve considerable performance, sparse spike activity, and fast inference. [ABSTRACT FROM AUTHOR]

The application of traditional deep learning methods for intelligent fault diagnosis is limited by the distribution discrepancy of the unlabeled data collected under different working conditions. Transfer learning can break through this limitation by generalizing a model trained on the source domain with massive labeled data to solve the fault diagnosis problem in the target domain with unlabeled data. The current transfer learning methods focus on directly measuring and minimizing the distribution discrepancy of the features between the two domains. These methods may confront difficulties when the distributions between the domains are complex and heterogeneous, and may cause the incorrect alignment of the same class data with the greatest distribution discrepancy across the two domains. In this paper, a deep transfer learning method with inter-domain decision discrepancy minimization (InDo-DDM) is proposed. The proposed method directly measures and minimizes the discrepancy of the decision result matrixes to facilitate the minimization of the distribution discrepancy between the two-domain data. With the proposed domain indicator, the InDo-DDM can find the greatest decision discrepancy and better align the data with the greatest distribution discrepancy. Additionally, the measurement of the decision discrepancy can be more precise and robust by introducing the nuclear-norm to avoid the fallible data classification near the decision boundary caused by the intra-batch imbalance. Extensive experiments in three different scenarios with two datasets from Case Western Reserve University (CWRU) and one dataset from Prognostic and Health Management (PHM) Data Challenge revealed that the InDo-DDM outperformed the other widely used methods. [ABSTRACT FROM AUTHOR]

Deep learning based methods are attractive and meaningful in the field of fault diagnosis recently. However, the sample size of different faults may be imbalanced in practical application scenarios, which results in performance degeneration. The optimization of data representation could be an effective way to alleviate such phenomenon. In this paper, a deep learning framework considering time series and distance metric called LSTM-Quadruplet Deep Metric Learning(LSTM-QDM) model is proposed. It maps original space to a feature space where the distribution of imbalance faults is more distinguishable. A novel quadruplet data pair is designed which adds a minor sample from imbalanced classes into traditional data pair. Based on such data pair, a quadruplet loss function is proposed to increase the distance between the imbalanced classes and other classes, and its combination with softmax loss would improve the representation ability and classification performance simultaneously. The proposed data pair and loss function encourage the full mining of imbalanced data in the training process. The experiments are carried out on two open-source datasets including TE and CWRU, and the fault diagnosis performance is validated under different imbalanced conditions. The experimental results indicate that our proposed model is effective and robust in the imbalanced fault diagnosis task. [ABSTRACT FROM AUTHOR]