Noninvasive prenatal screening using dynamic iterative depth optimization
Fetal maternal samples taken from pregnant women include both maternal cell-free DNA and fetal cell-free DNA. Described herein are methods for determining a chromosomal abnormality of a test chromosome or a portion thereof in a fetus by analyzing a test maternal sample of a woman carrying said fetus, wherein the test maternal sample comprises fetal cell-free DNA and maternal cell-free DNA. The chromosomal abnormality can be, for example, aneuploidy or the presence of a microdeletion. In some embodiments, the chromosomal abnormality is determined by measuring a dosage of the test chromosome or portion thereof in the test maternal sample, measuring a fetal fraction of cell-free DNA in the test maternal sample, and determining an initial value of likelihood that the test chromosome or the portion thereof in the fetal cell-free DNA is abnormal based on the measured dosage, an expected dosage of the test chromosome or portion thereof, and the measured fetal fraction.
1. A computer implemented method comprising:
sequencing a maternal sample of a woman carrying a fetus at a first sequencing depth of 6 million sequencing reads or more to obtain at least six million genetic sequence reads, and aggregating the sequence reads into bins that are each at least one kilobase each in length, the maternal sample comprising fetal cell-free DNA and maternal cell-free DNA;
generating, by a computer processor, a bin count vector for the sequence reads, the bin count vector comprising, for each bin, a count of sequence reads;
determining a fetal fraction in the maternal sample, wherein determining the fetal fraction in the maternal sample comprises inputting, by the computer processor, the bin vector into a machine learning model trained using bin count vectors and known fetal fractions corresponding to a plurality of training maternal samples;
measuring a dosage of a genomic region of the fetus, wherein measuring the dosage comprises determining, by the computer processor, an average number of sequencing reads per bin and a variation of the number of sequencing reads per bin in the genomic region;
determining, by the computer processor, based on the measured dosage, an expected dosage, and the fetal fraction, a value of likelihood of a chromosomal abnormality in a genomic region of the fetus and a first value of statistical significance;
identifying, by the computer processor, a no-call status, wherein identifying the no-call status comprises determining that (i) an absolute value of the first value of statistical significance is below a first threshold and (ii) the value of likelihood is above a second threshold;
re-sequencing, responsive to identifying the no-call status, the maternal sample at a second sequencing depth that is higher than the first sequencing depth;
determining, based at least on the re-sequencing, a second value of statistical significance;
identifying, by the computer processor, an abnormal status, wherein identifying the abnormal status comprises determining that an absolute value of the second value of statistical significance is above the first threshold; and
providing, by the computer processor, a report indicative of the abnormal status to at least one of the woman, a healthcare provider, or an institution.
2. The method of claim 1 , wherein measuring the dosage of the genomic region further comprises performing an assay to generate a plurality of quantifiable products, wherein a number of quantifiable products in the plurality of quantifiable products indicates the measured chromosome dosage of the genomic region.
3. The method of claim 2 , wherein the quantifiable products are sequencing reads or PCR products.
4. The method of claim 1 , further comprising:
computing, by the computer processor, a scalar factor that accounts for differences between fetal fractions in female pregnancies and fetal fractions in male pregnancies.
5. The method of claim 1 , wherein the plurality of training maternal samples comprises at least 100 training maternal samples.
6. The method of claim 1 , wherein the machine learning model is a regression model.
7. The method of claim 1 , wherein the machine learning model is a random-forest model.
8. The method of claim 1 , further comprising training the machine learning model, wherein training the machine learning model comprises applying a supervised machine learning technique to the bin count vectors and the known fetal fractions corresponding to the plurality of training maternal samples.
9. The method of claim 1 , wherein the genomic region corresponds to a first chromosome in the maternal sample, and wherein the expected chromosome dosage is determined by measuring a second average number of reads per bin and a second variation of the number of reads per bin for at least one chromosome other than the first chromosome in the maternal sample.
10. The method of claim 1 , wherein the genomic region corresponds to a first chromosome corresponding to the maternal sample, wherein the measured dosage is a first measured dosage, and wherein the expected dosage for the first chromosome is determined by measuring a second dosage of at least one chromosome other than the first chromosome from the maternal sample.
11. The method of claim 1 , wherein the genomic region corresponds to a first chromosome corresponding to the maternal sample, wherein the measured dosage is a first measured dosage, and wherein the expected dosage for the first chromosome is determined by measuring a second dosage for each chromosome of a plurality of chromosomes other than the first chromosome from the maternal sample, and determining, based on the second dosages, an average dosage for the plurality of chromosomes.
12. The method claim 1 , wherein the genomic region corresponds to a first chromosome corresponding to the maternal sample, wherein the machine learning model is a first machine learning model, wherein the measured dosage is a first measured dosage, and wherein the method further comprises determining the expected dosage for the first chromosome, wherein determining the expected dosage comprises:
i. generating a dosage distribution vector comprising a second measured dosage of at least one chromosome other than the first chromosome for each maternal sample in the plurality of training maternal samples;
ii. training a second machine-learning model by regressing the dosage distribution vector onto the first measured dosage for each maternal sample in the plurality of training maternal samples; and
iii. applying the second trained machine-learning model to a dosage distribution vector comprising the second measured dosage to obtain the expected dosage for the first chromosome in the maternal sample.