MosaicTR: tandem repeat somatic instability quantification from long-read sequencing

MosaicTR is a computational tool that leverages haplotype-tagged long-read sequencing to accurately quantify per-locus somatic instability of tandem repeats, overcoming the limitations of short-read methods to serve as a biomarker for mismatch repair deficiency and a monitor of disease progression in repeat expansion disorders.

The Gist

Imagine your DNA is a massive library of instruction manuals. Most of these manuals are written perfectly, but some sections contain a strange quirk: a specific phrase or sentence is repeated over and over again, like a stutter.

In a healthy person, these repetitions are stable. But in certain diseases (like Huntington's) or in cancer, these repetitions start to glitch. They might get longer or shorter randomly as cells divide. This is called somatic instability. It's like a photocopier that keeps adding or deleting a few words every time it copies a page, eventually making the instructions unreadable.

The problem is that scientists have struggled to measure exactly how much these glitches are happening, especially when looking at specific pages (loci) in the library.

Enter MosaicTR, a new digital tool described in this paper. Think of it as a super-smart, noise-canceling microscope for reading these glitchy DNA sections.

Here is how it works, broken down into simple concepts:

1. The "Long-Read" Advantage: Reading the Whole Sentence

Old tools used "short-read" sequencing. Imagine trying to understand a long, repetitive sentence by only reading 3 words at a time. If the sentence is "The cat sat on the mat on the mat on the mat," and you only see "on the mat," you have no idea how many times it repeats. You might guess, or you might get confused by the "stutter" of the machine (PCR stutter).

MosaicTR uses long-read sequencing (like PacBio and Oxford Nanopore). This is like reading the entire sentence in one go. You can see exactly how many times the phrase repeats, even if it's very long.

2. The "Haplotype" Trick: Sorting the Twins

Humans have two copies of every gene (one from mom, one from dad). Sometimes, one copy is healthy, and the other is glitching.

• The Problem: If you just mix all the DNA together, it's like having two different versions of a book in a pile. You can't tell which page belongs to which book.
• The Solution: MosaicTR uses "haplotype tags" (HP tags). Think of these as color-coded sticky notes. It separates the "Mom copy" from the "Dad copy" so it can measure the glitching on each one individually. This is crucial because often, only one side is breaking down.

3. The "Noise Filter": Ignoring the Static

This is the paper's biggest innovation. Long-read machines are great, but they aren't perfect. They sometimes make tiny, random mistakes (like adding an extra letter or missing one) that look like a glitch but aren't.

• The Analogy: Imagine trying to hear a whisper in a windy room. The wind (sequencing noise) makes it hard to know if the person is actually whispering a secret or if it's just the wind.
• The Fix: MosaicTR knows the "rhythm" of the DNA. The DNA repeats come in specific blocks (motifs), like a 3-letter word.
  • If the machine makes a mistake that breaks the 3-letter rhythm (e.g., adding 1 letter), MosaicTR knows, "Ah, that's just wind noise," and ignores it.
  • If the change is a whole block (e.g., adding a full 3-letter word), MosaicTR knows, "That's a real glitch!" and counts it.
    This "rhythm filter" makes the tool incredibly accurate, cutting out false alarms.

4. The "Instability Score" (HII)

The tool gives you a score called the Haplotype Instability Index (HII).

• Score near 0: The DNA is stable. It's a calm library.
• High Score: The DNA is chaotic. The repetitions are growing or shrinking wildly. This tells doctors that a disease might be starting or that a tumor has a specific type of DNA repair failure.

5. Time Travel and Map Making

The tool isn't just for a single snapshot. It can:

• Compare Time: Look at DNA from a patient at age 20 and again at age 40 to see how the glitches grew over time (longitudinal studies).
• Compare Places: Look at DNA from the brain vs. the liver to see if the glitches are happening everywhere or just in one specific organ (tissue-specific).

Why Does This Matter?

• For Disease: In diseases like Huntington's, the more the DNA glitches, the worse the symptoms. MosaicTR helps predict how fast a disease might progress.
• For Cancer: Some cancers have broken DNA repair systems. MosaicTR can spot these broken systems early, acting as a "check engine light" for cancer cells.
• For Research: It allows scientists to finally see the "mosaic" nature of our bodies—how different cells in the same person can have slightly different DNA instructions.

In short: MosaicTR is a new, high-precision ruler that can measure the tiny, chaotic changes in our DNA's repetitive sections, filtering out the background noise to give doctors and scientists a clear picture of disease progression and cancer risks.

Technical Summary

1. Problem Statement

Tandem repeat (TR) somatic instability—the variation in repeat length within an individual's cells over time or across tissues—is a critical factor in the onset and progression of repeat expansion disorders (e.g., Huntington's disease, Fragile X syndrome) and serves as a biomarker for mismatch repair deficiency in cancer.

Current methods face significant limitations:

• Short-read sequencing: Tools like prancSTR and MSIsensor are limited by read length (cannot span large expansions) and suffer from PCR stutter artifacts, particularly at trinucleotide repeats.
• Existing Long-read tools: While genotypers like TRGT and LongTR can size large alleles, they lack specific mechanisms to distinguish biological somatic instability from platform-specific sequencing noise.
• Lack of Haplotype Resolution: Most tools analyze pooled reads, failing to distinguish instability on one allele versus the other in heterozygous individuals, which is essential for accurate disease characterization.
• Noise Sensitivity: Different platforms (PacBio HiFi vs. Oxford Nanopore) have distinct error profiles (e.g., sub-motif homopolymer indels vs. random 1-bp jitter) that confound instability metrics if not specifically modeled.

2. Methodology

MosaicTR is a Python-based tool designed to quantify per-locus somatic instability from haplotype-tagged (HP-tagged) long-read BAM files.

Core Workflow

1. Input & Read Extraction:
   • Accepts HP-tagged BAM files and a BED catalog of loci.
   • Extracts reads with sufficient flanking sequence (≥50 bp) and high mapping quality.
   • Groups reads by haplotype using HP tags. If tags are missing, it employs gap-split analysis (bimodality test) or pooled analysis.
2. Allele Sizing:
   • Calculates allele sizes directly from CIGAR strings (sum of alignment operations spanning the repeat) without sequence realignment.
   • Computes median allele sizes per haplotype using MapQ-weighted medians with conditional motif-unit rounding.
3. Haplotype Instability Index (HII):
   • The primary metric quantifies intra-allelic variation normalized by motif length ($\ell$).
   • Formula: $HII_k = \frac{1}{\ell} \cdot \frac{\sum w_i |s_i - \bar{s}_k|}{\sum w_i}$
     • $s_i$: Read allele size.
     • $\bar{s}_k$: Haplotype median.
     • $w_i$: Weight based on motif-unit consistency.
   • Noise Reduction Strategy:
     • PacBio HiFi: 92% of errors are sub-motif homopolymer indels. MosaicTR down-weights deviations that are not integer multiples of the motif length ($w=0.1$) while giving full weight to whole-motif deviations ($w=1$).
     • Oxford Nanopore (ONT): Characterized by random $\pm1$ bp jitter. For motifs $\ge3$ bp, 1 bp is sub-motif (down-weighted). For dinucleotides, 2 bp equals one motif unit (full weight), requiring platform-specific noise thresholds (e.g., 2.0 for dinucleotides, decreasing to 0.5 for motifs $\ge7$ bp).
4. Asymmetry & Comparison:
   • Inter-haplotype Asymmetry Score (IAS): Measures if instability is confined to one allele ($IAS \approx 1$) or symmetric ($IAS \approx 0$).
   • Multi-sample Modes: Supports pairwise comparisons (e.g., tumor vs. normal) and matrix modes to track longitudinal changes ($\Delta HII$) across multiple tissues or time points.

3. Key Contributions

• Motif-Unit-Aware Noise Separation: A novel weighting scheme that distinguishes biological somatic expansion (whole motif units) from sequencing noise (sub-motif errors), significantly reducing false positives on both PacBio and ONT platforms.
• Haplotype-Resolved Instability: The first tool to quantify instability at the individual allele level without requiring additional phasing steps at each locus, enabling the detection of heterozygous carriers where instability is allele-specific.
• Multi-Platform Adaptability: Automatically detects sequencing platforms and applies specific noise models (e.g., adjusting thresholds for ONT dinucleotide repeats).
• Longitudinal & Tissue-Specific Analysis: Provides modes to detect tissue-specific instability patterns and track expansion accumulation over time (passage number).

4. Results

• Validation & Accuracy:
  • On synthetic data, HII scaled linearly with simulated instability ($R^2 = 1.0$).
  • Classification Performance: Achieved an AUC of 0.975 (99.3% sensitivity, 100% specificity) with a threshold of HII = 0.45.
  • Noise Baseline: In healthy HG002 samples (108,584 loci), the median HII was 0.004. Motif-unit weighting reduced the false positive rate from 10.2% to 0.9%.
  • Comparison: MosaicTR demonstrated a narrower noise distribution than Owl (which uses unweighted coefficient of variation).
• Disease Carrier Detection:
  • Successfully identified carriers for 5 disorders (Huntington, SCA10, Fragile X, SCA3, SCA1) across PacBio and ONT datasets.
  • Gradient Correlation: HII values correlated with expansion size (e.g., SCA10 carrier with 1,041 repeats had HII = 31.0).
  • Haplotype Resolution: Distinguished between unilateral instability (IAS $\approx$ 1) and bilateral expansion (IAS $\approx$ 0.12), a distinction impossible with pooled data.
• Longitudinal Tracking:
  • In the HG008 pancreatic cancer cell line, MosaicTR detected progressive expansions (e.g., +675 bp gain in a GAGCC repeat) between passages 23 and 41.
  • Observed that 90% of drifted loci were dinucleotide repeats, consistent with replication slippage, while mononucleotide instability was absent (consistent with microsatellite stability).

5. Significance

MosaicTR addresses a critical gap in genomic analysis by enabling the precise quantification of somatic mosaicism in tandem repeats using long-read sequencing. Its ability to separate sequencing noise from biological signal through motif-unit weighting makes it robust across different sequencing technologies.

The tool is particularly significant for:

1. Clinical Diagnostics: Improving the detection of repeat expansion disorders by distinguishing true somatic instability from technical artifacts and resolving heterozygous states.
2. Cancer Research: Providing a sensitive biomarker for mismatch repair deficiency and tracking tumor evolution.
3. Biological Insight: Facilitating large-scale studies (e.g., SMaHT consortium) to map tissue-specific and age-dependent somatic expansion patterns genome-wide, which was previously hindered by the limitations of short-read technologies and lack of haplotype resolution.

Availability: The tool is open-source (MIT license) and available at https://github.com/junsoopablo/mosaictr.