Journal: Genome Medicine
Article Title: Tumor-naïve ctDNA detection with deep learning-enhanced error suppression for sensitive mutation calling
doi: 10.1186/s13073-026-01694-y
Figure Lengend Snippet: Our combined approach can accurately predict ctDNA in CRC patients. A Characteristics of 32 preoperative plasma samples derived from 31 CRC patients. For more details, see Additional file 1: Table S3. B cfDNA levels in ng/ml blood for healthy and CRC preoperative samples from primary and metastasized patients. C Comparison of mutation calls when applying different stages of the pipeline across 32 CRC patients. The left plot shows the number tumor- confirmed mutations, and the right plot shows tumor- unconfirmed mutations (a proxy for false positive) D Fragment length distribution of tumor confirmed, tumor unconfirmed, and healthy control mutations in plasma. E Called mutations of DEEPctMUT and DEEPctMUT without PBMC on a cohort of 22 CRC patients and 27 healthy controls. F Called mutations of DEEPctMUT, DEEPctMUT without PBMC, Avenio, and Avenio BAM files analyzed by PI DeepES and RF on an independent cohort of 10 CRC patients and 10 healthy controls according to two VAF ranges. G Patient-level performance of the DEEPctMUT, DEEPctMUT without PBMC, Avenio, and Avenio analyzed by PI DeepES and RF on 10 CRC patients and 10 healthy controls. For Avenio (with iDES), a VAF threshold of > 0.5% was applied. For all other pipelines, VAF > 0.03% was applied. H VAF with DEEPctMUT of mutations in metastatic and localized tumor samples
Article Snippet: Synthetic cfDNA Pan-cancer Reference Standards were purchased from Twist Bioscience (WT and 5% VAF; 3.0 μg per tube [San Francisco, CA, US]).
Techniques: Clinical Proteomics, Derivative Assay, Comparison, Mutagenesis, Control