COMET assigns individual native long reads (PacBio HiFi or Oxford Nanopore) from bulk tissue to their cell type of origin, using the CpG methylation pattern each read carries and a cell-type reference methylation panel.
comet assignsplits each sample into cell-type read sets (BAM and/or per-site methylation BED) and reports and plots cell-type composition and cell-type-resolved genome-wide methylation.comet dmrcalls differentially methylated regions between two groups of per-site BED files with metilene, e.g. one cell type across donor groups or one cell type against the others.
Bioconda
conda create -n comet -c conda-forge -c bioconda comet-meth
conda activate cometFrom source
git clone https://github.com/fanglab/COMET.git
conda env create -f COMET/environment.yml && conda activate comet
export PATH=$PWD/COMET/bin:$PATHDownload the reference panel (hg38; Zenodo DOI:
10.5281/zenodo.24064938) and give it to COMET with --ref-panel:
mkdir -p ref_panel
curl -fL -o ref_panel/brain.atlas.bed.gz "https://zenodo.org/records/24064938/files/brain.atlas.bed.gz?download=1"
curl -fL -o ref_panel/brain.atlas.bed.gz.tbi "https://zenodo.org/records/24064938/files/brain.atlas.bed.gz.tbi?download=1"COMET is tested on Linux (x86-64). Bioconda and environment.yml install everything it needs:
| Software | Version | Used by |
|---|---|---|
| Python with numpy, pandas, scipy, pysam | Python ≥ 3.8, pysam ≥ 0.21 | comet assign |
| matplotlib, mscorefonts (Arial) | matplotlib ≥ 3.5 | comet assign visualization |
| modkit | 0.3.1 | comet assign ONT analysis |
| metilene, perl | 0.2.9 | comet dmr |
| samtools, bedtools, htslib (bgzip, tabix) | ≥ 1.15, ≥ 2.29, ≥ 1.15 | comet assign and comet dmr |
Test the installation with comet test. It downloads ~1.9 GB and takes ~10 min with
16 threads, including the download. It runs four steps:
- Checks the required tools and Python packages and prints their versions.
- Runs
comet dmron the test BEDs bundled with COMET. - Downloads the test BAMs and the brain reference panel (unless given with
--ref-panel) from Zenodo and checks their md5 sums. - Runs
comet assignon the test BAMs.
Every result is compared with the expected values and reported as PASS or FAIL.
comet test --env-only # step 1 only (seconds, no download)
comet test --threads 16 --out comet_test # all steps; downloads and outputs go to comet_test/# Assign cell type for PacBio HiFi brain data
comet assign --platform pacbio --ref-panel brain.atlas.bed.gz --bam sample.bam --assign-thres 0.65 --out output/sample/
# Assign cell type for ONT brain data
comet assign --platform ont --ref-panel brain.atlas.bed.gz --bam sample.bam --fasta hg38.fa --assign-thres 0.65 --out output/sample/
# Assign cell type for ONT brain data with additional modification calls (5hmC at CpGs)
comet assign --platform ont --ref-panel brain.atlas.bed.gz --bam sample.bam --fasta hg38.fa --assign-thres 0.65 \
--ont-mods 5mCG,5hmCG --out output/sample/--assign-thres sets the minimum confidence for a read to be assigned to a cell type.
Value recommended from our simulations: 0.65 for brain.
Run comet assign --help or comet dmr --help for full usage information.
Input: an aligned, coordinate-sorted and indexed modBAM with MM/ML tags (hg38, chr1–chr22):
PacBio HiFi CCS reads with the np tag, or ONT reads from dorado (plus the reference FASTA).
The cell-type reference panel of the tissue is given with --ref-panel (e.g. brain.atlas.bed.gz,
6 cell types).
Usage: one or several samples in one call.
comet assign --platform pacbio --ref-panel brain.atlas.bed.gz --assign-thres 0.65 \
--bam brain_01.bam,brain_02.bam,brain_03.bam \
--threads 16 --out outputOutput: each sample's results are written to output/<sample>/:
- Read assignment: per-read cell-type posteriors and assignments.
- Reads per cell type: BAMs and per-site methylation BEDs.
- Quantification tables: cell-type composition and genome-wide methylation of each cell type.
output/summary/ combines the tables and plots of all samples. For example, from three
postmortem brain samples:
Cell-type composition (summary/all_samples.composition.tsv, .pdf/png): proportions of
the reads assigned to a single cell type.
| sample | Exc_neuron | Inh_neuron | Astrocyte | Microglia | Oligodendrocyte | OPC |
|---|---|---|---|---|---|---|
| brain_01 | 0.127 | 0.137 | 0.101 | 0.289 | 0.273 | 0.073 |
| brain_02 | 0.212 | 0.157 | 0.086 | 0.311 | 0.173 | 0.062 |
| brain_03 | 0.286 | 0.145 | 0.078 | 0.309 | 0.132 | 0.050 |
Cell-type-resolved methylation (summary/all_samples.gw_methylation.tsv,
all_samples.cell_type_methylation.pdf/png): genome-wide CpG methylation of all reads (Bulk) and
of each cell type's reads.
| sample | Bulk | Exc_neuron | Inh_neuron | Astrocyte | Microglia | Oligodendrocyte | OPC |
|---|---|---|---|---|---|---|---|
| brain_01 | 0.795 | 0.758 | 0.839 | 0.753 | 0.737 | 0.716 | 0.739 |
| brain_02 | 0.815 | 0.763 | 0.842 | 0.776 | 0.752 | 0.728 | 0.757 |
| brain_03 | 0.796 | 0.737 | 0.832 | 0.767 | 0.746 | 0.712 | 0.742 |
Input: per-site methylation BED files (chrom, start, end, n_mod, n_total), e.g. the cell-type
BEDs written by comet assign. Give them as two groups: --target and --background.
Usage: with the example BEDs in COMET/example/data/dmr/:
d=COMET/example/data/dmr/brain_deep
comet dmr --target ${d}_AD_1.Microglia.CG.bed.gz,${d}_AD_2.Microglia.CG.bed.gz --target-name AD_Microglia \
--background ${d}_MCI_1.Microglia.CG.bed.gz,${d}_MCI_2.Microglia.CG.bed.gz,${d}_NCI_1.Microglia.CG.bed.gz,${d}_NCI_2.Microglia.CG.bed.gz \
--background-name MCI_NCI_Microglia --dmr-diff 0.2 --dmr-level individual --out output/dmrDMRs are called with metilene, then filtered by --dmr-diff, CpG count and depth, at group or
individual level (--dmr-level).
Output: output/dmr/01.dmr/ has every DMR metilene reported (*.all.tsv), with group means,
the methylation level of each input file (mean_<sample>), statistics and filter flags. The
significant DMRs are also written as *.sig.tsv and *.sig.bed.
For example: microglia of two AD samples against microglia of two MCI and two NCI samples, at
--dmr-level individual, so each region differs from all four by at least 0.2.
Two DMRs fall in chr15:74,174,956-74,175,254 (methylation columns are mean_<sample>; the target
comes first):
| chrom | start | end | n_cpg | AD_1 | AD_2 | MCI_1 | MCI_2 | NCI_1 | NCI_2 | min_individual_diff |
|---|---|---|---|---|---|---|---|---|---|---|
| chr15 | 74174956 | 74175055 | 9 | 0.147 | 0.114 | 0.879 | 0.727 | 0.663 | 0.590 | 0.443 |
| chr15 | 74175090 | 74175254 | 7 | 0.263 | 0.346 | 0.792 | 0.716 | 0.647 | 0.590 | 0.244 |
The six methylation columns show DMR methylation levels. The target (AD microglia) comes first.
The region is unmethylated in AD microglia and methylated in MCI and NCI microglia:
Plot any region from the comet dmr output with comet_plot_dmr.py:
comet_plot_dmr.py --dmr-dir output/dmr --region chr15:74,174,700-74,175,500 --out output/dmr/dmr_region

