Study On EMD Denoising Algorithm And Its Application In Colon Cancer Gene Expression Data

Currently, the research on gene expression dataset is one of the hotspots in the field of biological information, because it may be exist certain relevance between the occurrence of some external living organism disease and internal gene expression, so through the analysis of gene expression dataset that associated with disease, find out the probable causative genes and explore the pathogenesis, thereby open a new path for the prevention and control of disease. However the acquisition process of dataset is relatively complex and is vulnerable to the system or human disturbance that will become noise component that affect normal expression of gene data, which unfavourable to the further analysis and processing of dataset. In order to improve the accuracy of the data analysis late, it will ensure the facticity of the original gene data, so it has important realistic significance for gene expression dataset denoising.The denoising study theoretical basis in this paper is Empirical Mode Decomposition (EMD), which the biggest advantage is that the signal can be adaptive decomposition, therefore on the basis of EMD this article will set to study on the denoising method and apply it to denoising analysis of gene expression dataset. The EMD threshold denoising method is not only simple on principle but also more flexible on algorithm design, through the analysis and summary of the existing EMD threshold denoising method, the cut-off point selection of Intrinsic Mode Function (IMF) and noise standard deviation computation in threshold are two key steps, which the algorithm design will affect the final denoising effect of signal.In the traditional median-value estimation EMD denoising, it is more convenient and reasonable to use the characteristic of signal autocorrelation function for judgment the demarcation point of IMF component, but use median-value estimate IMF’noise standard deviation lack of pertinence, the accuracy of the results is difficult to guarantee, finally influence the denoising effect of signal. By contrast, the noise standard deviation computation based on white noise decomposition characteristics can better reflect white noise distribution characteristics in the IMFs under EMD, which is a new computation way, so this paper will take the median-value estimation EMD denoising as foundation, combined with new noise standard deviation computation so as to optimize denoising strategy. On the optimized scheme it is not only retain the advantage of IMF demarcation point selection in the traditional EMD denoising but also enhance the computation accuracy of IMF noise standard deviation, thereby improving the denoising result of signal.The above advanced denoising method has a better effect on white noise, but the method should be further enhanced to other types of noise on denoising result, aiming at the above shortage, this paper uses the noise standard deviation computation under the fractional Gaussian noise (fGn) model and makes corresponding improvement for applying it to EMD denoising, which as another optimization strategy of the median-value estimation EMD denoising. The computation method under fGn model can reflect the relationship of noise standard deviation of each IMF component, improved calculation method can further enhance the pertinence of this denoising, at the same time, the fGn is a generalized model of discrete gaussian white noise, so it can not only reduce common white noise but also decrease interference of colored noise, then strengthen the denoising effect.The proposed optimization denoising scheme in this paper will through the simulation signal and colon tumor gene expression dataset denoising experiment for testing and validation, results show that, scheme in the paper can retain more useful information in the denoising, denoising effect is superior to median-value estimation EMD denoising on the whole, meanwhile provide a reference scheme for denoising of the colon tumor gene expression dataset.