Motivation:Core fucosylation is a common type of glycosylation that plays a significant role in biological functions. Accurate identification of core fucosylated glycopeptides is challenging due to fucose migration phenomenon during mass spectrometry. By using glycopeptides from mouse brain with FUT8 knocked out as cases and core-fucosylated high-mannose glycans in normal mouse brain as controls, the phenomena are widely observed from mass spectrometry data. The relative intensities of 10 core-related characteristic ions are used jointly as a feature vector, and a semisupervised model and a self-supervised model are developed in the feature space with robustness of the models studied. Results:Experimental results show that both models perform well, with the former superior to the latter, reaching 99.95% identification accuracy on an independent mouse brain data with FUT8 knocked out. By applying the models to wild-type mouse brain, human IgG and human serum, their dominant abundance of core fucose and/or noncore fucose are found, which is trustworthy since the effect of fucose migration is dealt with. The study highlights the great significance of trustworthy data labeling, well-defined features, and machine learning/deep learning techniques in highly reliable, accurate, and robust identification of core fucose from high-throughput mass spectrometry data. Availability and implementation:The code for core fucose identification is freely available in https://github.com/yzy-010203/core_focuse_identification.

In recent years, tourism revenue and visitor numbers in Northwest China have increased steadily. However, many tourists still have limited knowledge of scenic destinations across the five northwestern provinces. When travelers intend to visit the region but have not yet decided on specific destinations, an intelligent recommendation system is urgently needed to assist their decision-making. Based on collaborative filtering, content matching, or knowledge graphs existing systems primarily face three major challenges: Due to reliance on historical data, the recommendation performance for new users and new attractions is weak; limited ability to capture tourists’ current intentions and personalized needs; insufficient utilization of multimodal information. To address these challenges, We propose a novel deep learning-based multimodal recommendation model, T-ECBM. A dataset comprising 23,488 user reviews and 4160 images of 52 attractions was collected. BERT was employed to extract semantic features from reviews, capturing subjective preferences and sentiment, while an improved EfficientNet-CA model extracted visual features from images to identify key scenic elements. The two feature sets were fused and fed into a multilayer perceptron, formulating the recommendation task as a multi-class classification problem. Experimental results demonstrate that text-only BERT achieved a Top-1 accuracy of 82.67%, while image-only EfficientNet-CA reached 83.68%. In contrast, the proposed T-ECBM achieved 96.71% Top-1 accuracy, 99.82% Top-5 accuracy, and an F1-score of 96.70%, proving its significant superiority over unimodal approaches. By integrating textual and visual modalities, T-ECBM effectively reduces information asymmetry, enriches decision-making support, and delivers intelligent, efficient, and personalized recommendations for tourists exploring northwestern China.

In recent years, contrastive learning has become a prominent approach in unsupervised learning. Based on its powerful feature extraction capabilities, contrastive learning can make full use of the value of a large amount of unlabeled high-quality data for model training, thereby improving model quality. However, contrastive learning still faces serious privacy leakage issues when used in downstream tasks. Although recent research attempted to address this issue by incorporating differential privacy into model training, these approaches still suffer from low model utility, especially in the scenarios of extracting accurate feature by contrastive learning. In this paper, we propose DP-CLMI, a novel methodology by integrating differential privacy into the feature output layer of the contrastive learning model which acts as a feature extractor and is then used by untrustworthy parties for downstream tasks, thereby safeguarding the privacy of the training data. Our experimental results demonstrate that this method not only maintains a robust guarantee of differential privacy, but also preserves acceptable levels of classification accuracy. Additionally, the methodology proves to be resilient against various forms of membership inference attacks, especially neural network-based MIA.

The backdoor attack has emerged as a prevalent threat that affects the effectiveness of machine learning models in intelligent vehicles. While such attacks may not impair the normal performance of the trained model, they can be exploited by malicious entities to manipulate model inferences, resulting in serious problems. In this paper, we design a dynamic gradient clipping (DGC) method aimed at rectifying backdoor models by eliminating the underlying backdoor trigger. Firstly, we construct a repair dataset fused by some clean samples and few-shot backdoor samples to amplify the backdoor behavior when we only obtain limited backdoor samples. Subsequently, we introduce sample states to characterize the backdoor behavior of the target model, determined by the model's inference outcome. Finally, we devise the DGC method to clip parameter gradients at varying degrees, effectively eliminating the backdoor trigger within the target model. Through the evaluation, the simulation results demonstrate that our DGC method exhibits robust defense capabilities against four contemporary state-of-the-art backdoor attacks, reducing the attack success rate by 95% with only 0.1% similar to 4.8% model accuracy loss.

Long non-coding RNAs (lncRNAs) play a crucial role in numbers of biological processes and have received wide attention during the past years. Since the rapid development of high-throughput transcriptome sequencing technologies (RNA-seq) lead to a large amount of RNA data, it is urgent to develop a fast and accurate coding potential predictor. Many computational methods have been proposed to address this issue, they usually exploit information on open reading frame (ORF), protein sequence, k-mer, evolutionary signatures, or homology. Despite the effectiveness of these approaches, there is still much room to improve. Indeed, none of these methods exploit the contextual information of RNA sequence, for example, k-mer features that counts the occurrence frequencies of continuous nucleotides (k-mer) in the whole RNA sequence cannot reflect local contextual information of each k-mer. In view of this shortcoming, here, we present a novel alignment-free method, CPPVec, which exploits the contextual information of RNA sequence for coding potential prediction for the first time, it can be easily implemented by distributed representation (e.g., doc2vec) of protein sequence translated from the longest ORF. The experimental findings demonstrate that CPPVec is an accurate coding potential predictor and significantly outperforms existing state-of-the-art methods.

Detecting the association between two variables is necessary and meaningful in the era of big data. There are many measures to detect the association between them, some detect linear association, e.g., simple and fast Pearson correlation coefficient, and others detect nonlinear association, e.g., computationally expensive and imprecise maximal information coefficient (MIC). In our study, we proposed a novel maximal association coefficient (MAC) based on the idea that any nonlinear association can be considered to be composed of some piecewise-linear ones, which detects linear or nonlinear association between two variables through Pearson coefficient. We conduct experiments on some simulation data, with the results show that the MAC has both generality and equitability. In addition, we also apply MAC method to two real datasets, the major-league baseball dataset from Baseball Prospectus and dataset of credit card clients' default, to detect the association strength of pairs of variables in these two datasets respectively. The experimental results show that the MAC can be used to detect the association between two variables, and it is computationally inexpensive and precise than MIC, which may be potentially important for follow-up data analysis and the conclusion of data analysis in the future.

In machine learning, when the number of available samples is very limited, it constitutes a small-sample problem. Small-sample problem is extremely challenging in learning a good model. As the first study on building the most reliable and generalizable model and providing the most reliable measure of a model's generalization performance, a large number of random sampling sets without or with noise added are created from observation dataset to simulate unseen training sets or unseen test sets respectively, where the former makes it available to consider constituting a model of the most generalization ability, and the latter makes it available to measure generalization performance of a model in a most reliable manner. In modeling process, each training set is used to train a model, and the models are then combined to form global/final model. In evaluation process, a performance metric (such as mean square error) is calculated on each test set, and the model's generalization ability is the most reliably represented by the average value of these metrics. The method is especially suitable to small or even ultra-small sample problems due to its no need of data splitting in cross-validation approaches. Comparative experiments on different datasets show the effectiveness of the proposed method in the case of ultra-small-sample problems: for simulated data, the generalization error is approximately 80% lower than those of the conventional method, and for a typical ultra-small-sample problem, the real data of glomerular filtration rate prediction, it is 46.36%∼83.02% lower than those of many popular models.

ABSTRACT Protein coding regions prediction is a very important but overlooked subtask for tasks such as prediction of complete gene structure, coding/noncoding RNA. Many machine learning methods have been proposed for this problem, they first encode a biological sequence into numerical values and then feed them into a classifier for final prediction. However, encoding schemes directly influence the classifier’s capability to capture coding features and how to choose a proper encoding scheme remains uncertain. Recently, we proposed a protein coding region prediction method in transcript sequences based on a bidirectional recurrent neural network with non-overlapping 3-mer feature, and achieved considerable improvement over existing methods, but there is still much room to improve the performance. First, 3-mer feature that counts the occurrence frequency of trinucleotides in a biological sequence only reflect local sequence order information between the most contiguous nucleotides, which loses almost all the global sequence order information. Second, kmer features of length k larger than three (e.g., hexamer) may also contain useful information. Based on the two points, we here present a deep learning framework with hybrid encoding for protein coding regions prediction in biological sequences, which effectively exploit global sequence order information, non-overlapping gapped kmer (gkm) features and statistical dependencies among coding labels. 3-fold cross-validation tests on human and mouse biological sequences demonstrate that our proposed method significantly outperforms existing state-of-the-art methods.

Precision mapping of glycans at structural and site-specific level is still one of the most challenging tasks in the glycobiology field. Here, we describe a modularization strategy for de novo interpretation of N-glycan structures on intact glycopeptides using tandem mass spectrometry. An algorithm named StrucGP is also developed to automate the interpretation process for large-scale analysis. By dividing an N-glycan into three modules and identifying each module using distinct patterns of Y ions or a combination of distinguishable B/Y ions, the method enables determination of detailed glycan structures on thousands of glycosites in mouse brain, which comprise four types of core structure and 17 branch structures with three glycan subtypes. Owing to the database-independent glycan mapping strategy, StrucGP also facilitates the identification of rare/new glycan structures. The approach will be greatly beneficial for in-depth structural and functional study of glycoproteins in the biomedical research.

考虑到网络用户数量的快速增长、日益复杂的网络环境以及网络应用程序多元化的现状,识别网络中的具体应用程序(诸如Google、Facebook、Skype、MSN等)是网络应用的重要研究方向,通过提取网络流量特征并利用机器学习方法识别网络应用程序是其主流方法,但由于网络流量特征多且复杂,经特征选择所获特征用于分类的性能往往严重依赖于所选用的分类器而不能很好反映应用程序的个性特征.为此本文提出了一种基于多样化组合特征选择的网络应用程序分类方法,通过组合特征重要性筛选和递归特征消除,获取这两种特征选择方法选择到的特征进行并操作,后用皮尔逊相关系数进一步去除冗余特征.对有87个特征的总计3577296个实例的网络数据集实验结果表明,与传统的诸如VT、RFE、L1正则化逻辑回归等特征选择方法相比,组合特征选择方法在KNN、SVM、RF、GBDT、XGBoost、LighGBM等各类分类器上均性能优异(分类准确率提升了0.5％-3.0％),且性能基本不受所采用的分类器的影响,表明所选出的特征能更客观反映网络应用程序的特性,同时所需运行时间也极大缩减(缩减了20％-90％),提升了网络应用程序实时监管的效率.

Translation initiation site prediction is crucial to understand the mechanisms of gene expression and regulation. Many computational approaches have been proposed and achieved acceptable prediction accuracy. Although recent Convolutional Neural Network-based method effectively learn consensus motifs and shows remarkable prediction performance, this method could not fully exploit coding features which have been proved significant to the identification of translation initiation sites. Indeed, coding features often exhibit higher-order distant interactions among nucleotides and learning this kind of feature from uncharacteristic mRNA sequences without any explicit biological knowledge is difficult. This situation gets worse when given no coding labels. In viewing of these shortcomings, we propose a novel method for translation initiation sites prediction in mRNA sequences based on a hybrid dependency network and deep learning framework (NeuroTIS) which explicitly model label dependencies among coding region, between coding region and translation initiation site. Meanwhile, a Bidirectional Recurrent Neural Network and a Convolutional Neural Network are employed for effective learning and inference. The experimental results show that the proposed framework yields an excellent prediction performance on two benchmark gene datasets, which significantly outperforms existing state-of-the-art methods.

Breast cancer is the most common malignancy in women, and because it has a high mortality rate, it is urgent to develop computational methods to increase the accuracy of breast cancer survival predictive models. Although multi-omics data such as gene expression have been extensively used in recent studies, the accurate prognosis of breast cancer remains a challenge. Somatic mutations are another important and promising data source for studying cancer development, and its effect on the prognosis of breast cancer remains to be further explored. Meanwhile, these omics datasets are high-dimensional and redundant. Therefore, we adopted multiple kernel learning (MKL) to efficiently integrate somatic mutation to currently molecular data including gene expression, copy number variation (CNV), methylation, and protein expression data for the prediction of breast cancer survival. Before integration, the maximum relevance minimum redundancy (mRMR) feature selection method was utilized to select features that present high relevance to survival and low redundancy among themselves for each type of data. The experimental results demonstrated that the proposed method achieved the most optimal performance and there was a remarkable improvement in the prediction performance when somatic mutations were included, indicating that somatic mutations are critical for improving breast cancer survival predictions. Moreover, mRMR was superior to other feature selection methods used in previous studies. Furthermore, MKL outperformed the other traditional classifiers in multi-omics data integration. Our analysis indicated that through employing promising omics data such as somatic mutations and harnessing the power of proper feature selection methods and effective integration frameworks, the breast cancer survival predictive accuracy can be further increased, thereby providing a more optimal clinical diagnosis and more effective treatment for breast cancer patients.

The next-generation sequencing technology offers a wealth of data resources for the detection of copy number variations (CNVs) at a high resolution. However, it is still challenging to correctly detect CNVs of different lengths. It is necessary to develop new CNV detection tools to meet this demand. In this work, we propose a new CNV detection method, called CBCNV, for the detection of CNVs of different lengths from whole genome sequencing data. CBCNV uses a clustering algorithm to divide the read depth segment profile, and assigns an abnormal score to each read depth segment. Based on the abnormal score profile, Tukey's fences method is adopted in CBCNV to forecast CNVs. The performance of the proposed method is evaluated on simulated data sets, and is compared with those of several existing methods. The experimental results prove that the performance of CBCNV is better than those of several existing methods. The proposed method is further tested and verified on real data sets, and the experimental results are found to be consistent with the simulation results. Therefore, the proposed method can be expected to become a routine tool in the analysis of CNVs from tumor-normal matched samples.

Translation initiation site (TIS) prediction is one of the most crucial subtasks for gene annotation. Many computational methods have been proposed and achieved acceptable accuracy in transcripts (e.g., cDNA, mRNA). However, the prediction of TIS at the genome level is far more challenging and the computational methods for TIS prediction in genomic sequences so far reach modest performance. Recently, we proposed a method that improves the prediction of TIS in mRNA sequences and demonstrated the significance of explicitly modeling coding features. In this paper, we extend the same results to genomic sequence and present a two-stage deep learning model for TIS prediction in genomic sequence: the first stage to extract coding contrast features around TIS by a hybrid Convolutional Neural Network-Bidirectional Recurrent Neural Network architecture (Content-RCNN), and the second stage to integrate coding contrast features around TIS with TIS sequence encoded by one-hot encoding to jointly predict TIS by a CNN (Integrated-CNN). Four-fold cross validation tests on genome-wide human and mouse datasets demonstrate that the proposed model yields an improved prediction performance of TIS over existing state-of-the-art methods. The source code and the dataset used in the paper are publicly available https://github.com/xdcwei/DeepTIS. (C) 2021 Published by Elsevier Inc.

Tonstein in coal seam (TCS) is a kind of mining solid waste, which was developed to a novel adsorbent (CuBr2-TCS) by using copper bromide modification. In this paper, CuBr2-TCS was subjected to elemental mercury (Hg0) removal experiment in simulated coal-fired flue gas (SFG). Several characterization methods were used to determine the mineralogical characteristics of TCS and reaction mechanisms. In-depth, the Hg0 removal performances of CuBr2-TCS under different flue gas components were explored. The results revealed that CuBr2-TCS exhibited 92.1% and 78.3% Hg0 removal efficiency in dry and wet SFG, respectively. HCl and O2 facilitated Hg0 removal performance of CuBr2-TCS by supplementing oxygen atoms and halogens, respectively, accompanying some intermediate transition products such as Cu2OBr2. SO2 played a serious suppressive role. SO2 acting alone or NO and SO2 acting simultaneously caused irreversible changes in the surface functional groups that formed active sites with NO. However, the thermal stability of the adsorbed mercury on the adsorbent which was spent in N2 + SO2 + O2 atmosphere became better. In addition, the spent adsorbent that first went through the Hg0 removal process in N2 + NO atmosphere, exhibited higher Hg0 removal efficiency in N2 + SO2 + NO atmosphere than that first reacted in N2 + SO2 atmosphere. CuBr2-TCS is a cost-effective adsorbent for the Hg0 abatement from the coal-fired flue gas (CFG).

While carbon capture, utilization, and storage (CCUS) is recognized as a highly promising technology for large-scale CO2 reduction from power and chemical industries, there have not been many investigative efforts on CCUS using biomass feedstock such as pepper stalk ash (PSA). Therefore, in this study, the direct carbonation of incinerated PSA via an aqueous route was studied under low-medium pressure conditions in an autoclave reactor. To optimize the reaction conditions in order to achieve the maximum carbonation efficiency, the complex effects of various operating parameters, such as the reaction temperature, liquid-solid (L/S) ratio, re -action time, pressure, and reaction gas, on the CO2 sequestration characteristics were systematically investigated. Under the theoretical framework of the shrinking core model (SCM) and assuming that the diffusion coefficient is a function of time, a new kinetic equation was proposed to provide a more precise fit to the experimental data of the heterogeneous direct carbonation. In addition to solid carbonates, PSA carbonation yielded dry residues from liquid products, which included 58.03 wt% potassium oxide and hence can potentially be used as feedstock for potassium fertilizers.

Comprehensive identification and cataloging of copy number variation (CNVs) are essential to providing a complete view of human genetic variation and to finding diseased genes. Due to the large-scale sequencing and cost control whole-genome sequencing (WGS) data, low-coverage data is favorably disposed towards CNV identification. However, such low-coverage data is sensitive to noise and sequencing biases, which results in low resolution of CNV detection in past experimental designs for WGS datasets. In this paper, we present a control-free Dirichlet process Gaussian mixture model (dpGMM) based approach, to analyze the read depth (RD) of low-coverage WGS datasets for CNV discovery. First, noise and biases of the RD signals are corrected through the preprocessing step of dpGMM. Then we assume that RD signals across genomic regions follow a Gaussian mixture model (GMM) in which each Gaussian distribution is followed by a copy number state. Without requiring the number of Gaussian distributions, dpGMM builds a Dirichlet process (DP) GMM for RD signals and further uses a DP prior to infer the number of Gaussian models. After that, we apply dpGMM to simulation datasets with different coverages and individual datasets, and compare ours to three widely used RD-based pipelines, CNVnator, GROM-RD, and BIC-seq2. Simulation results demonstrate that our approach, dpGMM, has a high F1 score in both low- and high- coverage sequences. Also, the number of overlaps between CNVs detected in real data by ours and the standard benchmark is twice as much as that detected by other tools such as CNVnator and GROM-RD.

Protein coding regions prediction is a very important but overlooked subtask for tasks such as prediction of complete gene structure, coding/noncoding RNA. Many machine learning methods have been proposed for this problem, they first encode a biological sequence into numerical values and then feed them into a classifier for final prediction. However, encoding schemes directly influence the classifier capability to capture coding features and how to choose a proper encoding scheme remains uncertain. Recently, we proposed a protein coding region prediction method in transcript sequences based on a bidirectional recurrent neural network with non-overlapping kmer, and achieved considerable improvement over existing methods, but there is still much room to improve the performance. In fact, kmer features that count the occurrence frequency of trinucleotides only reflect the local sequence order information between the most contiguous nucleotides, which loses almost all the global sequence order information. In viewing of the point, we here present a deep learning framework with hybrid encoding for protein coding regions prediction in biological sequences, which effectively exploiting global sequence order information, non-overlapping kmer features and statistical dependencies among coding labels. Evaluated on genomic and transcript sequences, our proposed method significantly outperforms existing state-of-the-art methods.

Characterizing copy number variations (CNVs) from sequenced genomes is a both feasible and cost-effective way to search for driver genes in cancer diagnosis. A number of existing algorithms for CNV detection only explored part of the features underlying sequence data and copy number structures, resulting in limited performance. Here, we describe CONDEL, a method for detecting CNVs from single tumor samples using high-throughput sequence data. CONDEL utilizes a novel statistic in combination with a peel-off scheme to assess the statistical significance of genome bins, and adopts a Bayesian approach to infer copy number gains, losses, and deletion zygosity based on statistical mixture models. We compare CONDEL to six peer methods on a large number of simulation datasets, showing improved performance in terms of true positive and false positive rates, and further validate CONDEL on three real datasets derived from the 1000 Genomes Project and the EGA archive. CONDEL obtained higher consistent results in comparison with other three single sample-based methods, and exclusively identified a number of CNVs that were previously associated with cancers. We conclude that CONDEL is a powerful tool for detecting copy number variations on single tumor samples even if these are sequenced at low-coverage.

Oxygen carriers (OCs) are the key factor in chemical looping combustion (CLC), a promising and efficient technology for converting coal and biomass into pure carbon dioxide and high concentrations of hydrogen. However, existing CLC techniques cannot completely separate OCs and ash, which could affect the performance of the OCs through the interaction with the ash-forming mineral matter at a high temperature. In this work, the effect of solid-solid carbothermal reactions on the Fe-based OC was systematically examined via changes in the carbon content and temperature. Furthermore, the CLC of the Fe-based OC was simulated by a fixed bed reactor, with oxidizing and reducing gases switched constantly. The results indicated that the added amount of biomass carbon had a significant influence on the Fe-based OC. Under highly reducing conditions, the reduction between Fe2O3 and C took precedence and inhibited the reaction between Fe2O3 and minerals. Under weakly reducing conditions, the reactions between Fe2O3 and minerals were promoted. In the looping experiment of the Fe-based OC and ash, a small amount of ash could promote the reduction of Fe2O3 . When ash accumulated, some foreign ions (Ca2+, K+, and Na+) in the ash interacted with the aluminosilicate to form compounds with lower melting points, covering the surface of Fe2O3 particles to hinder the reaction. Increasing the cycle temperature not only accelerated the reaction of Fe2O3 but also promoted the formation of large particles, such as (Fe2SiO4)(x)center dot(Fe3O4)(1-x )and CaMgSi2O6, which aggravated sintering and decreased the oxygen-releasing performance of the OC.