摘要
传统的药物理化分析鉴别方法普遍存在破环性检测、实时效力差以及制备流程复杂等问题。为此,可考虑使用拉曼光谱结合机器学习的方法进行快速建模分析。依据原研药和仿制药等谱图特征,结合机器学习及深度学习等算法,最终使用PSO-RF模型,针对不同批次的原研药片和仿制药片识别准确度达97.5%。同时,在对各类药物进行一致性评价量化时,通过基于傅里叶频域变换的余弦相似度FFT-Coisne及光谱信息散射度SID和光谱角SAM值量化了各类药品和标准原研药的相似度差异。
Traditional physical and chemical analysis identification methods generally suffer from problems such as destructive detection,poor real-time effectiveness,and complex preparation processes;Consider using Raman spectroscopy combined with machine learning methods for modeling and analysis.Considering the spectral characteristics of original drugs and generic drugs,combined with machine learning and deep learning algorithms,the PSO-RF model was ultimately used,and the accuracy of tablet recognition for different batches was 97.5%.At the same time,when quantifying the consistency evaluation of various drugs,the similarity differences between various drugs and standard original drugs were quantified using cosine similarity FFT Coisne based on Fourier frequency domain transformation,spectral information scattering degree SID,and spectral angle SAM values.
作者
于永爱
罗智嘉
邓赵斌
陈娟
Yu Yongai;Luo Zhijia;Deng Zhaobing;Chen Juan(college of Computer and Information Engineering,Shanghai Polytechnic University,Shanghai 201209,China;Shanghai Oceanhood Opto-electronics Tech Co.,Ltd.,Shanghai 201216,China)
出处
《实验与分析》
2024年第4期39-44,共6页
LABOR PRAXIS
关键词
机器学习
拉曼光谱
药物鉴别
量化一致性指标
Machine learning
Raman spectroscopy
Drug identification
Quantitative consistency indicators
作者简介
于永爱(1981年-),高级工程师,研究方向:光谱仪器研制,光谱数据分析,E-mail:eric.yu@oceanhood.com。
