2018/12/07 by Kabir Manghnani, Adam Drake, Manghnani, Kabir +5
Biochemistry, Genetics and Molecular Biology · #Bioinformatics and Genomic Networks #Cancer Genomics and Diagnostics #FOS: Computer and information sciences #Gene expression and cancer classification #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Single-cell and spatial transcriptomics
paper · pdf · doi:10.48550/arxiv.1812.03188
openalex publication_date 2018/12/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
High-dimensional data acquired from biological experiments such as next\ngeneration sequencing are subject to a number of confounding effects. These\neffects include both technical effects, such as variation across batches from\ninstrument noise or sample processing, or institution-specific differences in\nsample acquisition and physical handling, as well as biological effects arising\nfrom true but irrelevant differences in the biology of each sample, such as age\nbiases in diseases. Prior work has used linear methods to adjust for such batch\neffects. Here, we apply contrastive metric learning by a non-linear triplet\nnetwork to optimize the ability to distinguish biologically distinct sample\nclasses in the presence of irrelevant technical and biological variation. Using\nwhole-genome cell-free DNA data from 817 patients, we demonstrate that our\napproach, METric learning for Confounder Control (METCC), is able to match or\nexceed the classification performance achieved using a best-in-class linear\nmethod (HCP) or no normalization. Critically, results from METCC appear less\nconfounded by irrelevant technical variables like institution and batch than\nthose from other methods even without access to high quality metadata\ninformation required by many existing techniques; offering hope for improved\ngeneralization.\n