Strategies for data reduction in non-targeted screening analysis: The impact of sample variability for food safety applications

•Most observed compounds were shared between samples but varied in abundance.•The molecular variability was greater between brands/oat type than date of purchase.•Differential analysis successfully highlighted spiked compounds in food samples.•> 50% of features were reduced from a variety of oat products using in-house databases.•A combination of approaches was beneficial for samples with multiple ingredients. While analytical methods targeting specific compounds are critical for food safety, analytes excluded from the targeted list will not be identified. Non-targeted analyses (NTA) using LC/HR-MS complement these approaches by producing information-rich data sets where molecular formula can be generated for each detected compound; however, data mining can be labor intensive. Thus, we examined different NTA approaches to reduce the number of compounds needing further investigation, without relying on a suspect list or MS/MS database, both in single ingredient foods (i.e., oats) and more complex, oat-containing samples. We investigated inherent sample variability and utilized this information to build in-house databases for removing food compounds from sample data. While food databases were useful for data reduction, differential analysis was the most promising approach for single ingredient foods because it substantially reduced the number of features while retaining spiked QC compounds; however, a combination of approaches was necessary with greater sample complexity.