Sequencing data from MethylScan assay on plasma

Machine learning models in biomedical research are often hindered by demographic imbalances in clinical datasets, leading to biased predictions that disadvantage minority populations. Existing bias correction methods face limitations in handling the heterogeneity of biomedical data and the complexity of demographic influences. Here, we present DeBias, a computational framework for mitigating demographic biases in high-dimensional biomedical datasets. DeBias eliminates bias-associated subspaces within the entire feature space using control samples and enables global bias correction across all features, effectively preserving disease-specific signals while minimizing demographic distortions. To demonstrate its effectiveness, we applied DeBias to cfDNA methylation data for cancer detection. It achieved a >6-fold reduction in the number of features containing biases and outperformed existing methods in improving the cancer detection performance for minority populations. Our method represented a promising step toward developing machine learning models that are more accurate, reliable, and equitable for minority populations.

Type: Other
Archiver: European Genome-Phenome Archive (EGA)

1 Dataset

Click on a Dataset ID in the table below to learn more, and to find out who to contact about access to these data

Dataset ID	Description	Technology	Samples
EGAD00001015815	10ng from each cfDNA sample was used as the input material. For gDNA samples, 1.5-2ug each was first sonicated by Covaris R230 system (Woburn, MA). Next, size-selection with 0.8x-1.5x two-step Ampure XP beads was performed to enrich the 150-200bp fragments. 200ng of the size-selected gDNA was used as input material. The pre-capture library construction was performed with NEB UltraII DNA library prep kit for Illumina (Ipswich, MA) and the adapter was the methylated EM-seq adapter. Before the first PCR amplification, the gDNA samples were subjected to bisulfite conversion by QIAGEN Epitect kit, and cfDNA samples were subjected to enzymatic conversion by NEBNext enzymatic methyl-seq kit if the hybrid capture panel designed for converted DNA was used downstream. Twist panel hybrid capture was performed with their standard protocol (Twist Bioscience, South San Francisco, CA). The concentration of the post-capture library after the second PCR amplification was measured by Qubit 1xdsDNA HS kit (ThermoFisher, Waltham, MA). Their quality was examined by TBE-UREA PAGE and Bioanalyzer before sequencing. Libraries were then sequenced with 150bp paired-end reads on Illumina machines by Genewiz, Inc. (South Plainfield, NJ, USA). This dataset contains MethylScan data of the plasma samples from 248 noncancer individuals and 27 cancer patients (17 liver and 10 GI-tract cancers).	Illumina NovaSeq X	1