Methods for genome characterization
As described below, disclosed herein are methods of analyzing DNA methylation in cell-free DNA (cfDNA) and genomic DNA (gDNA) from sequencing data.
1. A computer-implemented method of characterizing the DNA methylation pattern in a biological sample, the method comprising:
(a) isolating fragments of DNA from a biological sample obtained from a subject, wherein the biological sample comprises cell free DNA (cfDNA) or genomic DNA (gDNA), wherein the gDNA has been contacted with an enzyme capable of cutting the gDNA at hypersensitive sites;
(b) constructing a library comprising said fragments, wherein the fragments are not treated with bisulfate;
(c) sequencing the library to less than 1x coverage to obtain sequence data;
(d) determining a predicted DNA methylation pattern of the cfDNA or gDNA in the sequence data using a Bayesian non-homogeneous Hidden Markov Model trained on training data comprising high coverage whole genome sequencing data using the parameters (i) fragment length of each individual DNA fragment, (ii) fragment coverage, and (iii) distance to fragment end.
2. The computer-implemented method of claim 1 , wherein the enzyme is Deoxyribonuclease I (DNase I).
3. The computer-implemented method of claim 1 , wherein the enzyme is Transposase.
4. The computer-implemented method of claim 1 , wherein the Transposase is TN5.
5. The computer-implemented method of claim 1 , wherein the sample comprises 1-20 ng of DNA.
6. The computer-implemented method of claim 1 , further comprising using the Bayesian non-homogeneous Hidden Markov Model to predict a methylation status at one or more CpG sites in cfDNA or gDNA.
7. The computer-implemented method of claim 6 , further comprising predicting a tissue of origin based on the predicted methylation status at the one or more CpG sites by comparing the predicted methylation status at the one or more CpG sites with a reference methylome.