Background: Liquid biopsies have been mainly used for cancer detection, non-invasive tumor genotyping and disease surveillance. Highly sensitive liquid biopsy assays have been developed that can now be applied to detect and characterize MRD. Several techniques have been developed for the detection of MRD, mainly by tracking ultra-low-frequency somatic tumor mutations in cell-free DNA. Methods: NGS using a 1123-gene-panel were performed on the tumor tissue and white blood cells which was designed for pan-cancer covering 1.8Mbp human genome. Moreover, we used a self-designed pan-cancer panel covering 208 kb and increasing 20 or less mutations individually known to be present in the surgical tissue based on the fixed panel for postoperative plasma or follow-up plasma. In Our study, plasma were collected from 137 healthy donors and UMI sequencing with 208 kb panel for built an database of cfDNA background noise. Furthermore, we constructed two different databases of CH(clonal hematopoietic) variants and negative-background noise with 2300+ cancer patients and 1000+ tumor-plasma paired data respectively. In this way, this approach not only could track selected tumor-specific mutations, but also provided with new variants using tumor-agnostic. MRD-positive status was defined as at least one variants in the plasma. Results: To test sensitivity and specificity of the minimal residual disease detection (MRD) method, we used 208kb panel data for plasma samples from 137 healthy samples and 150 samples with CRC untreatment. The result showed an area under the curve (AUC) of 0.95, with sensitivity of 92% and specificity of 98%. To test the limit of detection of 208 kb panel, we sequenced six different dilution levels (5%, 1%, 0.5%, 0.1%, 0.05% and 0.01%) with a positive control sample containing 98 known mutations. The result showed the detection limitation of 208 kb panel achieved 0.02%. We retrospectively analyzed 3000+ patients with colorectal cancer, 35 of whom met the eligibility criteria were enrolled in this study: 1) With stage II-III CRC undergoing radical surgery; 2)Blood specimens were collected on postoperative 7 days to 3 months; 3)Follow-up duration was 2 years or relapsed; 4)Enough samples for targeting sequencing. Twelve (34%) patients with MRD+,which all patients recurred while 23(66%) patients were MRD-, which 22/23 patient did not experience local recurrence or metastasis. The Kaplan-Meier estimates showed that the positivity of ctDNA was of high predictive value for relapse (HR 21, 95%CI 18.3-28.9;P < 0.001). Conclusions: In this study, combination of tumor-informed and tumor-agnostic strategies can effectively distinguish healthy and cancer patients. Meanwhile,our study highlights that postoperative ctDNA status can effectively detect MRD and identify patients with a high risk of relapse.