5. Row-wise Aggregation of Methylation Sites

yame rowop provides fast, low-level operations applied row-by-row across all samples of a .cx file.
These operations are frequently used in single-cell methylation analysis, pseudobulk aggregation, QC, and exploratory co-methylation analyses.

Usage: yame rowop -o <operation> <in.cx> <out>

If <out> is omitted, results are written to stdout.

---

5.1 Common Use Case: Summing Binary Data to Create Pseudobulks

A very common operation is to sum binarized methylation calls across cells belonging to the same cluster.

For example, to generate a pseudobulk for cluster 1 from a single-cell .cg file:

yame subset -l cluster_1_id.txt single_cell.cg \
  | yame rowop -o binasum - > cluster1_pseudobulk.cg

What binasum does:

• For binary formats (0/1): It counts how many cells have methylation = 1 vs 0.
• For format 3 (M/U counts): It compares M vs U in each cell and votes for the majority state. Only rows with coverage ≥ -c mincov are counted.

The output is a format 3 .cg file where each position contains:

• Total number of methylated votes (M)
• Total number of unmethylated votes (U)

This is ideal for downstream:

• Pseudobulk methylation beta estimates
• Differential methylation analysis
• Cluster-level QC

---

5.2 Available Operations

Below is a complete list of operations supported by -o.

---

1. binasum — Sum binarized methylation across samples

yame rowop -o binasum input.cx output.cx

• Works on formats 0, 1, and 3
• Produces a new format 3 dataset
• For format 3 inputs, rows with depth < -c (default 1) are ignored
• Interprets each sample’s methylation as a binary vote

Useful for:

• Pseudobulk construction
• Collapsing replicates
• Aggregating barcodes / technical partitions

---

2. musum — Sum M and U counts directly

yame rowop -o musum input.cx output.cx

• Works only on format 3

• Directly adds up:

  • all M counts
  • all U counts
• Produces a new .cx file in format 3

Useful when:

• You want true aggregated read counts
• You trust the original M/U values (not just binary interpretation)

---

3. mean — Per-row methylation mean and count

yame rowop -o mean input.cx > mean.tsv

Output (tab-delimited):

beta_mean    number_of_valid_cells

Only samples with coverage ≥ -c contribute. Rows with no valid values output NA 0.

Useful for:

• QC summaries
• Computing single-cell methylation averages
• Identifying globally variable CpGs

---

4. std — Per-row methylation standard deviation

yame rowop -o std input.cx > std.tsv

Output columns:

std_dev    counts

Uses the same filtering as mean. Helpful for:

• Measuring variability
• Ranking variable CpGs
• Identifying DNA methylation landmarks

---

5. binstring — Convert methylation profiles to binary strings

yame rowop -o binstring -b 0.6 input.cx > binstrings.txt

• Converts each sample to a binary string per row (one line per row, one character per sample)
• Threshold -b decides methylated vs unmethylated (default 0.5)

Example output:

0010110101
1100001110
...

Cell calls

For each sample at a row:

• 1 — confident methylated (beta > -b)
• 0 — confident unmethylated (beta < -b)
• ambiguous — no/low depth (mu == 0 or cov < -c) or an exact tie (beta == -b)

Ambiguous cells are not guessed at random. They are filled deterministically with the row’s majority confident state (more 0s → 0, more 1s → 1; exact tie → 0). Output is therefore reproducible, and -s no longer affects binstring.

Trusting the fill: sweeping majority (-M)

Filling from a near-even split (e.g. 6 vs 5) fabricates unreliable calls, so a row is only filled when its majority is sweeping: the larger confident side must be at least -M <fold> times the smaller (unanimous counts as sweeping; default -M 10). If a row has ambiguous cells but the majority is below this fold, the whole line is emitted as the 2 sentinel (below). Rows with no ambiguous cells always keep their real 0/1 calls, regardless of how thin the split is.

Filtering and the 2 sentinel (-m)

A row’s line is emitted as an all-2 string — same length as every other line, so positional alignment with the input row universe is preserved — when either:

• its ambiguous fraction exceeds -m <frac> (default 1.0 = no filtering), or
• it has ambiguous cells but no sweeping majority to fill them (see -M).

2 never appears in an otherwise-valid line, so downstream tools can treat any line containing 2 as a filtered row.

Option	Meaning	Default
-b <beta>	methylated if beta > beta	0.5
-c <mincov>	cells with cov < mincov treated as ambiguous	1
-m <frac>	max ambiguous fraction before the row is sentinel-filtered	1.0 (off)
-M <fold>	min majority fold (hi/lo of confident calls) to trust the fill	10

Useful for:

• Clustering by Hamming distance
• Haplotype inference
• Co-methylation motif discovery
• Compression for machine learning models

---

6. cometh — Co-methylation of neighboring CpGs

yame rowop -o cometh -w 5 input.cx > cometh.tsv

For each CpG i and neighbors i+1 … i+w, cometh outputs a vector encoding four categories:

• U0U1
• U0M1
• M0U1
• M0M1

Counts are stored in a compact 64-bit representation; use -v to print them explicitly:

i   U0U1-U0M1-M0U1-M0M1   U0U2-U0M2-M0U2-M0M2   ...

Additional behavior:

• Only format 3 is allowed
• Excludes intermediate methylation values (0.3–0.7)
• Requires both CpGs to have depth ≥ -c mincov
• -w sets window size (default: 5)

This is useful for:

• Detecting locally coordinated methylation
• Assessing haplotype methylation patterns
• Fragment-level epiallele analysis
• Identifying footprints or nucleosome-scale structure

---

5.3 Summary of Operations

Operation	Output Type	Input Requirement	Purpose
binasum	.cx (format 3)	fmt 0/1/3	Pseudobulk / aggregation
musum	.cx (format 3)	fmt 3	True count summation
mean	text	fmt 3	Mean methylation per CpG
std	text	fmt 3	Standard deviation per CpG
binstring	text	fmt 3	Binary methylation strings
cometh	text	fmt 3	Local co-methylation patterns

---

5.4 Additional Notes

Minimum coverage control (-c)

Many operations ignore rows where:

M + U < mincov

Default is 1.

Verbose output (-v)

For cometh, verbose mode expands packed category counts into readable numbers.

Binarization threshold (-b)

Relevant only for binstring.

---

5.5 Help and Subcommand Documentation

For detailed usage:

yame rowop -h