Introduction
Tandem repeats (TRs)—sequences where a specific DNA pattern repeats consecutively—comprise approximately 3% of the human genome. These regions are critical evolutionary hotspots and are frequently implicated in genetic diseases and gene silencing. However, analyzing their history is challenging due to the interplay between two competing forces: tandem duplication, which increases sequence length, and substitution, which introduces variations that make repeats "imperfect."
Traditional methods for estimating mutation rates often rely on cross-species comparisons (e.g., human vs. chimpanzee) or computationally intensive combinatorial algorithms. These approaches are either data-dependent or prohibitively slow for large-scale genomic analysis. This paper introduces the Stochastic Model of Tandem Repeats (SMTR), a framework designed to infer evolutionary history from a single genome with high efficiency.
The SMTR Framework
Instead of treating mutation history as a discrete combinatorial puzzle, SMTR models sequence evolution as a continuous-time dynamic system using stochastic approximation. The core of the method lies in analyzing the autocorrelation function of the DNA sequence, which captures the rhythmic periodicity of the repeats.
By comparing the observed autocorrelation of a real sequence to the theoretical limits predicted by an Ordinary Differential Equation (ODE) model, the researchers employ a least-squares curve-fitting approach. This allows for the estimation of substitution probabilities versus duplications of varying lengths. Because the TR effectively contains "approximate copies" of its own past states, the model can treat internal variations as a historical record, enabling single-genome inference.
Validation and Performance
The SMTR model was validated using Monte Carlo simulations to generate synthetic sequences with known histories. The results demonstrated that the model is not only accurate but also exceptionally fast. SMTR achieves a 1,000-fold speedup over existing state-of-the-art combinatorial methods like DTSCORE. For instance, processing a sequence with 120 mutations takes approximately 0.015 seconds with SMTR, compared to 15 seconds with previous algorithms.
As shown in Figure 1, the estimation error for both the mutation probability vector ($\hat{\mathbf{q}}$) and the total mutation count ($\hat{n}$) decreases significantly as the number of mutations increases. This confirms the model's robustness, particularly for longer, more established repeat sequences.
Biological Insights from the Human Genome
Applying SMTR to the human genome (HG38) revealed several key biological insights:
- Natural Clustering: TRs naturally divide into two distinct groups based on mutation profiles. One group is dominated by duplications (microsatellites), while the other is dominated by substitutions (minisatellites).
- Genomic Proxies: Mutation activity within TR regions correlates strongly with the overall mutation rate of the host chromosome. SMTR independently identified the same high-mutation chromosomes (Y, 21, 22, 19, and 16) previously found only through complex cross-species genomic comparisons.
- The GCD Rule: A mathematical finding of the study is that the observed period of a tandem repeat converges to the greatest common divisor (gcd) of all possible duplication lengths, rather than just the most frequent one.
Conclusion and Future Directions
SMTR transforms a heavy computational problem into a light statistical one, providing a scalable tool for genomic research. While highly effective, the current model has limitations: it does not yet account for deletions (indels), biased substitutions (e.g., GC-content bias), or repeats formed by unequal recombination. Future iterations of the model aim to incorporate these complex biological factors to further refine our understanding of genomic instability and evolution.
Summary updated