A study of structural determinants in the interleukin-1 fold

The structures of interleukin-1ß, basic fibroblastgrowth factor and Erythrina trypsin inhibitor have been analysedin order to determine whether the hydrophobic core remains conserved,even when the structures have extremely low sequence similarities.We find that there are significant differences in the way eachprotein achieves a satisfactory arrangement of core residuesand that positions which contribute to the core of one structureare not guaranteed to contribute to the integrity of another.Furthermore, the side-chain packing arrangements of these coreresidues vary significantly between the three structures. Duringthis analysis the side-chain rotamers for three independentlydetermined interleukin-1ß structures were also compared.It was found that although buried residues are generally inagreement the remaining residues frequently occupy differentrotamers in the three structures. This suggests that althoughmeaningful studies are possible for buried side-chains the resultsobtained from equivalent analyses of accessible residues shouldbe treated with caution. These results are discussed with specificreference to the optimization of side-chain packing in proteinsof known structure.     

应用于临床生存时间预测的二次学习算法研究

生存时间预测在医学、经济和工程等领域有着广泛的应用。随着机器学习技术和数据挖掘技术的发展和广泛应用,研究人员提出了很多基于机器学习技术的生存时间预测算法。这些算法虽然都取得了良好的效果,但预测精度均有提升的空间。因此,提出了一种基于二次学习风范的生存时间预测算法,并结合最近邻算法在截尾样本估计上的应用以及支持向量机在泛化性能上的优势,实现了对临床生存时间的建模。实验结果表明,该算法能够获取精确的生存时间,且具有预测精度上的性能优势。     

A predicted three-dimensional structure of human cytochrome P450: implications for substrate specificity

A three-dimensional structure for human cytochrome P450IA1 waspredicted based on the crystal coordinates of cytochrome P450camfrom Pseudomonas putida. As there was only 15% residue identitybetween the two enzymes, additional information was used toestablish an accurate sequence alignment that is a prerequisitefor model building. Twelve representative eukaryotic sequenceswere aligned and a net prediction of secondary structure wasmatched against the known -helices and ß-sheets ofP450cam. The cam secondary structure provided a fixed main-chainframework onto which loops of appropriate length from the humanP450IA1 structure were added. The model-built structure of thehuman cytochrome conformed to the requirements for the segregationof polar and nonpolar residues between the core and the surface.The first 44 residues of human cytochrome P450 could not bebuilt into the model and sequence analysis suggested that residues1–26 formed a single membrane-spanning segment. Examinationof the sequences of cytochrome P450s from distinct gene familiessuggested specific residues that could account for the differencesin substrate specificity. A major substrate for P450IA1, 3-methyl-cholanthrene,was fitted into the proposed active site and this planar aromaticmolecule could be accommodated into the available cavity. Residuesthat are likely to interact with the haem were identified. Thesequence similarity between 59 eukaryotic enzymes was representedas a dendrogram that in general clustered according to genefamily. Until a crystallographic structure is available, thismodel-building study identifies potential residues in cytochromeP450s important in the function of these enzymes and these residuesare candidates for site-directed mutagenesis.     

基于深度网络的出租车Pick-up需求预测

在智慧城市建设中,区域的出租车需求预测是一个十分重要的模块。为了预测指定区域未来时刻的出租车需求,本文通过拓展已有序列模型,提出一个多时间分辨率的基于层次注意力机制的循环高速网络（Multi-Time Resolution Hierarchical Attention-Based Recurrent Highway Networks, MTR-HRHN）。MTR-HRHN将对外生数据时空特征的提取和目标数据的时空建模整合到单个框架中,并且通过多分辨率（例如每个小时或者每天）对序列数据不同的时间属性建模,从而捕获更全面的时间模式。最后,在纽约市出租车数据集上评估MTR-HRHN的预测性能。实验结果表明,与其他经典时间序列预测方法相比,MTR-HRHN在多个高需求区域的短期需求预测上表现出更好的预测性能。     

A novel deep learning prediction model for concrete dam displacements using interpretable mixed attention mechanism

Dam displacements can effectively reflect its operational status, and thus establishing a reliable displacement prediction model is important for dam health monitoring. The majority of the existing data-driven models, however, focus on static regression relationships, which cannot capture the long-term temporal dependencies and adaptively select the most relevant influencing factors to perform predictions. Moreover, the emerging modeling tools such as machine learning (ML) and deep learning (DL) are mostly black-box models, which makes their physical interpretation challenging and greatly limits their practical engineering applications. To address these issues, this paper proposes an interpretable mixed attention mechanism long short-term memory (MAM-LSTM) model based on an encoder-decoder architecture, which is formulated in two stages. In the encoder stage, a factor attention mechanism is developed to adaptively select the highly influential factors at each time step by referring to the previous hidden state. In the decoder stage, a temporal attention mechanism is introduced to properly extract the key time segments by identifying the relevant hidden states across all the time steps. For interpretation purpose, our emphasis is placed on the quantification and visualization of factor and temporal attention weights. Finally, the effectiveness of the proposed model is verified using monitoring data collected from a real-world dam, where its accuracy is compared to a classical statistical model, conventional ML models, and homogeneous DL models. The comparison demonstrates that the MAM-LSTM model outperforms the other models in most cases. Furthermore, the interpretation of global attention weights confirms the physical rationality of our attention-based model. This work addresses the research gap in interpretable artificial intelligence for dam displacement prediction and delivers a model with both high-accuracy and interpretability.     

基于数据挖掘的航空装备故障成品预测模型

因航空装备系统复杂度高,故障成因复杂,在保障时根据故障现象判断故障成品的难度很大,同时大量的历史故障记录数据未能有效利用.通过分析航空装备成品故障记录的特点,提出了一种基于数据挖掘的航空装备故障成品预测模型.该模型将历史故障记录数据作为输入,通过文本聚类将大量故障描述聚类得到故障现象簇,并建立"故障现象"与"故障成品"之间的多对多关系,提出了故障成品概率分布算法,并通过匹配新发生故障现象与故障现象簇,计算出故障成品概率分布.实验验证结果表明,该模型可有效地根据故障描述预测可能发生故障的成品的概率分布,且预测准确率可随数据量的增加而提高,满足实际保障需求.     

基于国产处理器增量式实时交通流预测算法及实现

针对城市交通难以处理大量数据且实时性差等问题,提出了根据增量式城市交通流数据预测拥堵情况的一种基于国产处理器的L-BFGS(limited-memory BFGS)算法。该算法通过存储向量序列计算Hessian矩阵,改进Two-Loop算法求下降方向,在Spark集群中并行处理时收敛速度快,适用于实时性要求强的城市交通场景。实验结果证明,L-BFGS预测算法完全可以在国产平台上对大规模的实时交通数据流进行快速建模、预测,在改善城市交通管理水平提供有效支撑的同时也丰富了国产芯片的应用领域。     

基于改进AdaBoost算法的复杂网络链路预测

链路预测是复杂网络的重要研究方向,当前的链路预测算法因可利用的网络信息有限,导致预测算法的精确度受限。为了提高预测算法的性能,采用改进的AdaBoost算法进行链路预测。首先根据复杂网络样本建立邻接矩阵,完成样本的矩阵化处理;然后采用AdaBoost算法进行分类训练,通过权重投票获取预测结果;最后,考虑到复杂网络弱分类器预测正负误差分布的不均衡问题,设置权重调整因子η及其调整范围[η1,η2],并根据η值动态调整AdaBoost算法的多个弱分类器分类结果的权重,从而获得准确的链路预测结果。实验结果证明,相比其他常用网络链路预测算法及传统AdaBoost算法,改进的AdaBoost算法的预测准确率优势明显,且在节点数量较多时,其预测时间性能和其他算法的差距较小。     

基于宽度学习的结构沉降预测

结构监测是确保工程结构建设在施工和运营阶段安全的关键因素,因此采用合理有效的预测模型对结构沉降监测数据进行科学准确的预测成为了当前结构沉降预测研究的重点。针对传统预测方法与深度学习方法用于结构沉降预测存在的预测精度不够高、模型结构复杂、训练耗时等问题,提出了一种基于宽度学习的结构沉降时间序列预测模型。通过实测地铁地下隧道沉降监测数据对宽度学习、人工神经网络、支持向量回归和深度置信网络-支持向量回归预测模型的预测结果进行对比分析。实验结果表明:宽度学习系统(broad learning system,BLS)应用于结构沉降预测具有良好的效果,其训练速度更快,预测精度更高。验证了所提出的宽度学习算法应用于结构沉降预测的可实施性和有效性。