Non-targeted GC, LC/MS instruments are capable of detecting very low concentrations of thousands of metabolomic features and are an important tool for drug discovery, for establishing mechanisms of action and for discovery of novel biomarkers. In large cohorts, samples are analyzed in batches often over prolonged periods of time and are prone to instrument drifts, and other technical variation in experimental conditions. Sometimes, design of the experiment can eliminate drifts, but when it is not possible, metabolites' concentrations need to be corrected for systematic drifts. Correction should be conservative and eliminate drifts while keeping the true signal intact. We developed a novel signal-correction method for large non-targeted metabolomic studies that combines testing for presence of drifts using ensemble of white noise tests and then applies higher levels of drift correction such as smoothing splines to calculate drift correction. Results were validated in independent validation pooled plasma dataset.