人工神经网络用于FTMS准确质量的测定

Application of Artificial Neural Network in Mass Calibration of FTMS

FTMS中离子质量和测量频率之间的关系通常采用回归法实现经验公式的描述,由于离子频率的测定受许多因素影响,离子质量和测定频率之间的准确关系很难以一种公式来描述。采用无隐含层的BP网络建立了测定频率和质荷比的数学关系,提高了离子质量预测的准确度,并成功地在内标法和外标法准确质量测定中应用,对阿奇霉素的降解产物质量测定的相对误差均小于2×10-6。

The multiple regression is often used in mass calibration in Fourier transform mass spectrometry （FTMS）. Because observed frequency can shift caused by collisional damping and ion space charge effect, it is difficult to express the relationship between ion mass and observed frequency in FTMS. The relationship was investigated by neural net works. The prediction performance of the calibration models constructed by multiple rer gression and neural networks were compared, and no hidden layer BP with 4-1 is superior on robustness of prediction ions mass. Artificial neural network approach can Provide better prediction results. The relative error of ions mass are less than 2×10^-6.