nanopore-methylation

By adapting EM algorithm in a Hidden Markov Model (HMM), we are able to cluster nanopore reads into their parental haplotypes. One can choose use all nanopore data from the chromosome or reads mapped to imprinted regions. By analysing nanopore reads raw signals, there should be one haplotype group shows where methylation happened. And we are able to located methylated bases.

Handle NA12878 data

imprinted_regions = read_imprinted_data("data/ip_gene_pos.txt")  
SNPS = load_VCF("data/chr19.vcf")  
READS = load_sam_file("data/chr19_merged.sam", "19", SNPS) # specify chromosome number

Find READS overlapping with any human imprinted region

o = get_overlapped_reads(READS, imprinted_regions)

Save pre-processed data

save_objects("data/snps.obj", SNPS)
save_objects("data/READS.obj", READS)
save_objects("data/reads_ir.obj", o)

Load pre-processed data

all_snps = load_objects("../data/snps.obj")
reads = load_objects("../data/READS.obj")
reads_ir = load_objects("data/reads_ir.obj")

HMM clusters nanopore reads into 2 possible haplotypes

Simulation

dummy_reads = load_objects("../data/dummy/dr1.obj")
dummy_snps = load_objects("../data/dummy/ds1.obj")
dm1 = run_model(dummy_snps, dummy_reads, 10)
dm2 = run_model(dummy_snps, dummy_reads, 10)
compare_models(dm1, dm2)

Run model

# Update all SNP sites
model = run_model(all_snps, reads, iter_num=100) # can choose to run model iter_num times
# Update part of SNP sites
model = run_model(all_snps, reads, 100, updateAll=False, p=0.5) # randomly choose 50% SNP sites to update

# Result: haplotypes (dict): {"m1": {snp_id: allele}, "m2": {snp_id: allele}}
haplotypes = model.get_haplotypes()  # each SNP site returns one allele

# Other results:
# Alleles clusters saved in a dictionary {"m1": {snp_id: [allele1, allele2]}, "m2": {snp_id: [allele1, allele2]}}
alleles_dict = model1.get_alleles()
sorted_alleles_dict = model1.get_ordered_alleles() # sorted by SNP positions
# haplotype strings, "-" on unassigned positions
h1, h2 = model1.align_alleles()

# Use trained model to cluster new nanopore reads data set (traning and new data set must from the same chr)
haplotypes_dict = model1.cluster(new_reads)

# Compare model grouping results
model2 = run_model(all_snps, reads, iter_num=100)
compare_models(model1, model2) # return common/different alleles and their positions

# Compare results of two trained models clustering new data set
compare_clustering(model1, model2, new_data)

YaoLi3/nanopore-methylation