Benchmarking NIPT algorithms on detecting numerical chromosome trisomy

Abstract

Noninvasive prenatal test (NIPT) is a widely used screening method to detect trisomy on chromosomes 13, 18, and 21. The lack of positive samples prevents us from examining the performance of NIPT algorithms on detecting trisomy on other chromosomes. Recently, we have introduced an efficient computational method to generate positive samples with trisomy from negative samples. In this paper, we applied the simulation method to generate 4600 positive samples for all 22 autosomal chromosomes as well as the X chromosome in females; and reused 1250 negative samples to assess the performance of algorithms CNVKit, WisecondorX, and VINIPT in detecting numerical chromosome aberrations. Experiments showed that WisecondorX had a sensitivity of 99.95% and a specificity of 97.2% on determining trisomy aberrations. VINIPT could detect all positive samples (i.e., sensitivity of 100%) and correctly determined 99.4% negative samples (i.e., specificity of 99.4%). The CNVkit algorithm was not as accurate as the WisecondorX and VINIPT algorithms. Its performance on some chromosomes such as chromosome 19 needs to be improved. WisecondorX and VINIPT could serve as reliable tools for analyzing NIPT data.