MIBD: Refining Demographic Inference through Structured Coalescent Theory

TL;DR

Understanding how populations move and mix (migration) is central to genetics. While Identity-by-Descent (IBD) segments are powerful tools for this, existing methods (like PIBD) often falter by ignoring lineages' "pre-coalescence" history. This paper introduces MIBD, a Markov-process-based approach that corrects these theoretical deviations, providing significantly more accurate estimates of population size and migration for small or ancient populations.

The "Independent Movement" Fallacy

In population genetics, the Island Model describes subpopulations exchanging migrants. Previous SOTA methods, such as Palamara and Pe’er’s PIBD, estimated the probability of two individuals sharing an IBD segment by calculating the probability they are in the same location at a given time.

However, they made a critical simplification: they treated the migration paths of two individuals as independent. In reality, if we are looking for the Most Recent Common Ancestor (MRCA), we must account for the fact that the two lineages must not have coalesced at any point prior to that time. Ignoring this "non-coalescence" history leads to a mathematical paradox where the probability distribution of time does not sum to one, and the model fails the "merge test"—where high migration should logically result in two populations behaving like one large one.

Methodology: The Markovian Jump

To solve this, the authors utilize a continuous-time Markov jump process. Instead of tracking individuals independently, they track the configuration of ancestors in different populations as a single state.

The Formalism

The state of the system is defined by $(i,j)$, representing the number of ancestors in population 1 and 2. The transitions involve:

• Coalescence: Two ancestors in the same population merge into one.
• Migration: An ancestor jumps from one population to another.

By applying Kolmogorov’s backward equations, the authors derived closed-form solutions for the cumulative distribution function of the tMRCA ($F(t)$).

Figure: The system of ordinary differential equations used to solve for the tMRCA distribution.

Why MIBD Wins: Theoretical and Experimental Evidence

1. The Limiting Behavior Integrity

One of the most elegant proofs in the paper is the "Infinite Migration" limit. As the migration rate $Mo\infty$, MIBD correctly converges to the IBD sharing formula of a single population with size $2N_e$. The prior method, PIBD, fails this test, yielding an overestimation that contradicts biological intuition.

2. Superiority in Small Populations

The error in simpler models is most pronounced when the effective population size ($N_e$) is small. In simulations using the ms simulator, MIBD's relative error remained near zero, while PIBD deviated by as much as 30-40%.

Figure 1: Distribution of tMRCA. Note how PIBD (red/blue) can exceed the theoretical bounds or deviate significantly from the MIBD (black) and simulation results.

Table 1: Quantifying the Average Relative Error. MIBD consistently outperforms PIBD across different IBD length thresholds ($u$) and population sizes ($N_e$).

Case Study: Palestinian and Bedouin Populations

When applied to real-world SNP data from the HGDP-CEPH database, both MIBD and PIBD yielded similar results ($N_e\approx 10,000$, $m\approx 0.001$). This confirms that for large, modern human stagnant populations, the simplified models are "good enough." However, MIBD remains the only theoretically sound choice for studying endangered species, historical bottlenecks, or highly mobile sub-groups where $N_e$ is small.

Conclusion and Future Outlook

MIBD fixes a fundamental gap in haplotype-based demographic inference. While the current implementation focuses on the symmetric island model, the framework of using Markov jump processes opens the door for more complex scenarios, such as:

• Asymmetric Migration: Populations with "source-sink" dynamics.
• Admixture Events: Precisely dating when two distinct groups merged.
• Ancient DNA: Utilizing shorter IBD segments (which represent more ancient history) with higher confidence.

As our ability to detect shorter IBD segments (under 1cM) improves with better phasing algorithms, the robustness of MIBD will become increasingly critical for peering deeper into our evolutionary past.