基于改进的自组织神经网络的基因剪切位点的识别

Recognition of the splice sites based on improved self-organizing feature maps

为提高基因序列中剪切位点的识别率,将无先导卡尔曼滤波器(UKF)和自组织神经网络(SOFM)相结合,给出一种非线性高维数据的聚类算法。利用无先导变换(UT)参数化SOFM邻域宽度函数的均值和方差,并采用UKF进行预测,完成SOFM参数的自适应过程。该算法用于基因剪切位点的识别结果表明:较SOFM与EKF参数自适应方法,该算法识别精度较高,验证了其有效性和可行性。

A clustering method for large quantities of high-dimensional data which combining unscented Kalman filter （UKF） with self-organizing feature maps （SOFM） was proposed to improve the recognition accuracy of splice sites among the gene sequences. The mean and variance of width of the neighborhood function were parameterized by unscented transform （UT） and then predicted by UKF to complete adaptive process of SOFM parameters. Tests on recognizing gene splice sites show that the proposed method has higher recognition accuracy comparing with SOFM and EFK-based parameter selfadaptive methods, which verifies its validity and feasibility.