小波分析和人工神经网络在生态学研究中的应用

墨西哥帽子小波和Morlet小波在生态格局分析中的应用 本研究采用Monte Carlo方法，探讨了对小波分析的格局进行统计显著性检验的普遍方法。为了更好利用小波分析和了解两个常用小波--墨西哥帽子小波和Morlet小波在格局分析中的优缺点，用生态学研究中常见的4个模拟格局和东灵山辽东栎林的样带数据对这两个常用小波的特性进行了分析和比较。研究结果表明：墨西哥帽子小波能较好地分析样带中的斑块和间隙以及它们的位置信息，Morlet小波能较好地分析样带中尺度及其位置信息。不同的小波通常在尺度分析和斑块和间隙分析之间有平衡，所以最好的方法是结合两种小波的优点。小波分析在处理生态数据时，受所使用小波本身特性的制约。用墨西哥帽子小波进行格局分析时，小波能量谱的等值图上不同格局所对应的峰有可能相互重叠，当所分析的多尺度格局的规模差别不大时，所分析格局规模所对应的峰可能相互融合。这些小波能量谱的等值图上不同格局所对应峰的重叠和融合可能会导致格局分析量图上多个峰的相互融合和屏蔽。所以在使用小波分析做格局研究时，也应尽可能地结合小波能量谱和量图上的信息，以得到较全面和精确的格局分析信息。本研究的结论能为小波分析的应用提供指导。 应用二维小波分析对暖温带阔叶林辽东栎更新格局的研究 本研究介绍了一种二维网格空间数据分析方法一二维小波分析。该方法不仅能分析格局的等级结构，而且也能得到所分析结构的位置信息。小波系数等值图上不同格局规模的斑块和间隙可直接和不同尺度上的生态过程和生境条件相联系。小波方差从二维小波分析导出，小波方差可将四维的小波系数降至二维的小波方差函数，并量化所分析格局规模对整个格局的贡献。本研究用三个模拟格局分析了二维小波的特性及二维墨西哥帽子小波和Halo小波的特性。因为自身的特性，Halo小波比墨西哥帽子小波能提供更高的分辨率。本研究也将二维小波分析应用于暖温带阔叶林的辽东栎更新格局研究中，分析的结果表明：辽东栎的更新发生在辽东栎成树斑块和林窗斑块重叠区域。 用交互验证和独立验证来测试神人工经网络模拟水稻分檗动态的泛化能力 人工神经网络不是基于对所模拟过程的理解，而是依赖于对所分析数据的内部结构。所以人工神经网络通常被认为是经验模型而不能外推，而且在训练数据和验证数据的范围之外肯定不能精确地预测所模拟的过程。本研究通过对交互验证和独立验证神经网络模型性能的比较，测试了神经网络模型在预测水稻分檗动态方面的泛化能力。同时，也对几种提高神经网络泛化能力的技术进行了比较。研究结果表明：在训练数据的变量范围内，神经网络在预测水稻分檗动态方面具有泛化能力。较少的训练数据样本导致了对训练数据过度吻合的和不具泛化能力的神经网络。要能使神经网络在预测水稻分檗动态方面具有泛化能力，训练数据的样本量至少9倍于神经网络的权值数目。当神经网络有多个输入变量且训练数据不足以保证神经网络的泛化能力时，建议在训练之前，采用主成分分析、对应分析及类似技术压缩输入变量的个数。在压缩输入变量的个数之后，如果训练数据的样本量还不足以保证神经网络的泛化能力，就应采用提高神经网络泛化能力的技木，如：jittering和强制训练停止等，特别是神经网络与机理模型的复合模型。因为神经网络的泛化能力问题具有普遍性，所以我们的研究结论不只是适用于水稻分檗动态的预测，也适用于其它的农业和生态神经网络模型。

mRNA 很可能携带三维遗传信息

The forming mechanism of the three - dimensional structures of proteins，i.e.the mechanism of protein folding，is a basic problem in molecular biology which is still unsolved unitl now. In which a core problem is whether there is the three – dimensional genetic information that decide the three - dimensional structures of proteins. However, the research on this field has mot yet been reported. Recently，we made a comparative study on the folded structures of more than 70 mature messeneger RNAs (mRNAs) and the three - dimensional structures of the proteins encoded by them，it has been found that there exist marked correspondences between their featured structures in the following aspects: 1.The number of the structural units. An RNA molecule can form a secondary structure(stem and loop structure) by the folding and the base pairing of itself. The elementary structural unit of an RNA secondary structure is hairpin(or compound hair pin).The regular structural unit in the secondary structure of a protein is # alpha # - helix or #beta# - sheet . We have found that the hairpin number in the secondary structure of each mature mRNA is equal or approximately equal to the number of the regular secondary structural unis of the encoded protein. 2 .Turning region. Turn is a main structrual element in the secondary structure of a protein, which decides the backbone orientation of a protein molecule to some extent .Our analysis shows that the nucleotide sequence segments in an mRNA which encode the turns of the corresponding protein are overall situated in the turning regions of the mRNA secondary structure such as haipin，bulge loop or multibaranch loops. 3 .The arrangement of structural elements in space. In order to understand the backbone orientation of an RNA molecule and the arangement of its structural elements in space，we have modeled the three一dimensional structure of the mRNA molecule on SGI workstation based on its secondary structure.The result shows that the spatial arrangement of most of the nucleotide sequence segments encoding the structural elements of a protein is consistent with that of these stretural exements in the protein. For instance，the nucleotide sequences corresponding to each pleated sheet of a # beta # - sheet structure are close to each other in the mRNA secondary stucture and in the three - dimensional structure，although some of the nucleotide segments are far apart from each other in the one - dimensional sequence. For another instance，the two triplet codons of cysteines which form a disulphide bridge geneal1y are very close to each other in the mRNA folded structure. In addition，we also analyzed the locations of the codons proline - coding and the distrbution of the nucleotide sequences #alpha# - helix - coding in the folded structures of mRNAs . Some distribution laws have been found. All of these results suggest that the transfer of the genetic information from mRNA to protein not only is one – dimensional but also is three - dime ns ional. That is，there exists the genetic information that decide the three - dimensional structures of proteins. To a certain extent，we could say that the mRNA folding detemines the protein folding. Based on these results，it would be possible to predict the three - dimensional structures of proteins from the primary，secondary and tertiary structures of the m RNAs at a higher accuracy.And more important is that a new clue has been provided to uncover the“spatial coding" of the genetic information.