Fetal Fraction and Statistical Power in Noninvasive Prenatal screening

Seminar Series

Tuesday, May 14, 2019 - 12:00
MSRBIII Seminar Room

Abstract:   In the noninvasive prenatal screen (NIPS) the cell-free DNA (cfDNA) in maternal plasma is sequenced.  As some cfDNA circulating in the plasma of a pregnant woman is from her fetus, NIPS can detect some genetic abnormalities in the fetus. For example, fetal trisomy can be detected by estimating a chromosomal dosage from sequencing reads and comparing it against dosages of euploid controls by the conventional Z-test.  The power of NIPS depends critically on percent of maternal cfDNA that is of fetal origin, the fetal fraction (FF).   FF can be estimated from Y chromosomes for male fetuses, but such an estimate becomes inaccurate when the fetus has abnormal copies of sex chromosomal such as XYY, and more importantly, it cannot be applied to female fetuses. Here we present a method that can reliably estimate FFs for both male and female fetuses by examining SNP markers across autosome. 
Even at very low sequencing depth, there are sizable number of SNPs that are covered by more than one read. This can be deduced by assuming Poisson distribution at each locus and can be verified by real data.  At those SNPs covered by at least two reads, we define read heterozygosity, and demonstrate that the read heterozygosity is a function of FF, which allows it to be inferred. By taking into account other factors that affect read heterozygosity, such as the maternal inbreeding coefficient and sequencing error, our method can be made to be both robust and accurate.  Our FF estimator can then be used as an informative prior in computing a Bayes factor to test for aneuploidy.  Our Bayesian approach is more powerful than the conventional Z-test method for two reasons: First, the test statistic is heavy-tailed under the null, which is not captured in normal distribution commonly assumed for the Z-test. Our Bayesian approach explicitly accounts for this non-normality through its prior specification. Second, the Z-test only compares a test static to the null, while a Bayesian method also compares it to the alternative (which is FF for fetal trisomy). This informative alternative prior reduces false positive and consequently increases power.  

We show that our method is very effective in inferring FF through mixing reads, mixing genomic DNAs and reanalysis of clinical samples. Through simulation, we demonstrate that our Bayesian method incorporating the knowledge of FF substantially increases power of NIPS, particularly for difficult problems such as micro-deletion and micro-duplication. In analysis of clinical samples, we were able to identify female-male twins thanks to the accurate FF inference, and we learned that there are about 5% in vitro fertilization (IVF) pregnancies in our clinical samples. With FF known, the efficacy of the NIPS can be greatly improved in the following aspects. First, it brings a powerful Bayesian method to screen for aneuploidy. Second, it allows us to declare ``no call" for small FFs to reduce false negatives. Third, screening for aneuploidy of sex chromosomes, such as XYY syndrome, becomes a simple problem. Lastly, it allows us to develop an adaptive design to sequence at a higher depth for samples with small FFs.

Yongtao (Grant) Guan, PhD
Assistant Professor
Department of Biostatistics and Bioinformatics