Dynamically selecting sequencing subregions for cancer classification

Abstract

Methods and systems for segmenting sequencing regions obtained from a sample interval are disclosed. sample contamination detection are disclosed. In particular, an analytics system accesses test sequences from a sample. The test sequences each include a sequencing region which, in aggregate, form an aggregate sequencing region. The analytics system segments sequencing regions from the aggregate sequencing region into sequencing subregions. Several methods of segmenting sequencing regions into sequencing subregions are disclosed: (1) maximizing cancer vs. non-cancer methylation beta differences, (2) minimizing cancer vs. non-cancer methylation beta differences, (3) segmentation based on CpG density in regions, (4) dynamic generation of sequencing subregions based on mutual information scores and cancer classification propensity. The analytics system applies selects sequencing subregions and applies a cancer classifier to those subregions to identify cancer presence in the sample.