Research on the application of dynamic weighted KNN with preprocessing based on a normal distribution in metabolomics...

In the field of metabolomics data analysis, missing values are a common challenge. Traditional k-nearest neighbors (KNN) imputation methods often overlook the distribution of the original data, resulting in suboptimal outcomes when addressing missing values in metabolomics. To better restore the data distribution and enhance the imputation results, this paper introduces a dynamic weighted KNN imputation algorithm with preprocessing based on the normal distribution (NDW-KNN). Initially, the similarity distance between samples is calculated to assign an appropriate k value to each sample. Subsequently, missing values are categorized based on the similarity of the neighbors of the target sample and undergo normal distribution preprocessing. Finally, an inverse distance weighting method is used to assign weights to each sample, thereby predicting missing values. Experimental results show that NDW-KNN achieved the best performance across three benchmark metabolomics datasets, reducing the average NRMSE and MAPE by 21.7 % and 32.9 % compared with traditional KNN, and by 4.5 % and 13.8 % compared with NS-KNN. Even under a missing rate as high as 30 %, NDW-KNN maintained the lowest imputation error and the highest consistency with the original data distribution, while exhibiting stronger intergroup discrimination in principal component analysis, demonstrating its excellent robustness and practical applicability.