AlphaSplitter

Ab initio discovery of satellite DNA monomer alphabets from tandem repeat arrays.

AlphaSplitter identifies periodic conserved anchor sites in satellite DNA, assembles them into chain templates, and classifies monomers by structural indels — without reference HMMs, curated databases, or prior knowledge of the satellite family.

Quick Start

AlphaSplitter is a single alphasplitter CLI with subcommands. Build once, then invoke.

# Build
cargo build --release          # or: make

# End-to-end pipeline (discover → cut → annotate)
./target/release/alphasplitter run arrays.10kb.fasta -o results/ -t 96

# Or stage-by-stage
./target/release/alphasplitter discover arrays.10kb.fasta -o chains.json -t 96
./target/release/alphasplitter cut      arrays.10kb.fasta -m chains.json -o monomers.tsv -t 96
./target/release/alphasplitter annotate monomers.tsv annotated.tsv

Run alphasplitter --help for the full subcommand list; every subcommand takes --help as well. make run INPUT=arrays.10kb.fasta OUTDIR=results THREADS=96 is the Makefile shortcut for the end-to-end path.

What it does

Input: FASTA of satellite DNA arrays (e.g., from Satellome or TRASH).

Output (under -o <outdir>):

• chains.json — per-SF motif bundles with claimed array membership
• monomers.tsv — one row per monomer: family, array_id, coords, letter, subtype, site_order, site_structure, sequence
• monomers_hor.tsv — HOR strings per array
• annotated.tsv — monomers.tsv + CENP-B columns
• families.json — per-family alphabet report
• family_consensus.fa — one consensus per (family, letter)
• letters/<family>/<letter>.tsv — all monomers of each letter, one row per monomer (sequence first, then metadata) — ready to feed into muscle/hmmbuild

Pipeline

satellome .10kb.fasta
  │
  ▼
alphasplitter discover      (per-period, iterative SF discovery)
  ├── Scan 8-mers → filter by periodicity + support + complexity
  ├── Grow by Shannon entropy (cap 11bp)
  ├── Discover links (exact co-occurrence, stable spacing)
  ├── Assemble chain → enrichment (pairwise distances + occupancy ≥30%)
  └── → chains.json (families[]: motifs + claimed array whitelist)
  │
  ▼
alphasplitter cut -m chains.json
  ├── Partition arrays by family (each array gets cut with only its
  │     family's motifs — no cross-SF contamination)
  ├── Scan both strands, strand-normalize
  ├── Cut at first-site positions → monomers
  ├── Classify per family: letter (indel-defined) + subtype (site variants)
  ├── site_order: abcde, ab-de, abdebcde (dimers)
  ├── site_structure: 0:a18b24c43d18e:57
  └── → monomers.tsv, monomers_hor.tsv, letters/<family>/<letter>.tsv,
       family_consensus.fa, families.json
  │
  ▼
alphasplitter annotate
  ├── CENP-B box scoring (nTTCGnnnnAnnCGGGn, both strands)
  ├── 9-char and 17-char CIGAR notation
  └── → annotated.tsv

Core Model

A satellite monomer is a chain of conserved anchor sites separated by variable spacers:

[--site_A--]---spacer---[--site_B--]---spacer---[--site_C--]...
  conserved    LENGTH      conserved    LENGTH     conserved
  sequence     matters!    sequence     matters!   sequence

Letter = defined by indels only (which sites are present/deleted + spacer length changes), NOT by substitutions. Letters are assigned locally within each SF family: "A" in P171SF0 is not the same object as "A" in P171SF1; the family column disambiguates.

Site presence detection: a site is present if the distance between its neighbors matches expected — no fuzzy sequence matching needed.

Subcommands

Subcommand	Purpose
run	End-to-end: discover → cut → annotate
discover	Chain-first anchor discovery + enrichment + multi-SF/period
cut	Array → monomers, letter/subtype classification, per-letter dumps
annotate	Add CENP-B box columns to a monomers TSV
find-box	Search for any degenerate box pattern (CENP-B, TIGD4, etc.)
spacing	CENP-B box spacing analysis at single-base resolution
find-periodic-boxes	De novo periodic box discovery from paired k-mers
reads {build-hmms,scan,classify,extract,alphabet}	ONT read processing (HPC-HMM workflow)
dev {find-phase,motif-graph}	Hidden research subcommands

Validated on

Species	Satellite	Period	Sites	Arrays	Monomers	Letters
Human (CHM13)	Alpha satellite	171bp	12	522/1223	411K	585
Human (CHM13)	Human sat I	2420bp	87	70	—	—
Mouse (C57BL/6J)	Minor satellite	120bp	8	17/28	44K	105
Zebra finch	191bp satellite	191bp	11	262	104K	279
Zebra finch	717bp satellite	717bp	33	70	82K	127
Zebra finch	3772bp satellite	3772bp	131	108	—	—
7 primate species	Alpha satellite	171bp	5 (v1)	—	8M	26

Alexandrov validation: 38/39 SF types recovered, 96.3% of monomers assigned.

Key Findings

• 585 structural letter types in human alpha satellite (15× more than HumAS-HMMER's 39)
• CENP-B variable positions are chromosome-specific suprachromosomal family markers
• CENP-B spacing within CDR = chromosome-specific nucleosome positioning signature
• Mouse centromeric satellite saturated with CENP-B boxes (89% of monomers)
• Zebra finch has two alternative box systems (BoxA-D at 355bp, TIGD4 at 715bp) on different chromosomes
• Convergent architecture: human (340bp), ZF (355bp), mouse (120bp) → 1 box ≈ 1 nucleosome repeat

Output Format (annotated TSV)

Manifest header (one # block from cut, one from annotate):

#alphasplitter v0.2.0 / cut
#input: arrays.10kb.fasta
#motifs_source: chains.json
#family P171SF0 period=171 sites: M1=ACATCACAAAG M2=AGAATGCTTCT ...
#columns: family array_id strand monomer_idx start end length letter subtype
#          site_order site_structure sites distances sequence
#alphasplitter v0.2.0 / annotate
#CENP-B_box_ref: nTTCGnnnnAnnCGGGn (17bp, searched on BOTH strands)
#cenpb_labels: B+ (score>=7), B? (score=6), B- (score<=5)

Columns:

Column	Description
family	SF family that cut this monomer (e.g. P171SF0)
array_id	Original FASTA name (no _rc suffix)
strand	+ kept forward, - array was revcomped during strand normalization
start, end	Coordinates within the array
letter	Structural type (indel-defined), local to family
subtype	Letter + site sequence variant
site_order	Chain sites: abcde, ab-de, abdebcde
site_structure	gap_before:sites_with_distances:gap_after
cenpb	B+ (≥7/9), B? (6/9), B- (≤5/9)
cenpb_cigar17	Full 17-char CIGAR
sequence	Monomer DNA sequence

Requirements

• Rust 1.70+
• Input: satellite arrays in FASTA (recommend ≥10kb arrays)

Citation

Komissarov A. (2026). AlphaSplitter: Ab initio discovery of satellite DNA monomer alphabets from tandem repeat arrays. In preparation.

License

MIT