[Paper Review] Haploid-Diploid Evolution: Nature's Memetic Algorithm

Michail‐Antisthenis Tsompanas, Larry Bull+2 more|arXiv (Cornell University)|Nov 13, 2019

Evolution and Genetic Dynamics4 citations

TL;DR

This paper proposes a novel memetic algorithm, Haploid-Diploid Evolutionary Algorithm (HDEA), inspired by eukaryotic life cycles to exploit the Baldwin effect through diploid representation and recombination. By evaluating fitness as a composite of two haploid genomes, HDEA improves search performance on rugged fitness landscapes, outperforming standard haploid evolutionary algorithms in both abstract NK model tests and PhysiCell simulations for optimizing nanoparticle-based cancer therapy.

ABSTRACT

This paper uses a recent explanation for the fundamental haploid-diploid lifecycle of eukaryotic organisms to present a new memetic algorithm that differs from all previous known work using diploid representations. A form of the Baldwin effect has been identified as inherent to the evolutionary mechanisms of eukaryotes and a simplified version is presented here which maintains such behaviour. Using a well-known abstract tuneable model, it is shown that varying fitness landscape ruggedness varies the benefit of haploid-diploid algorithms. Moreover, the methodology is applied to optimise the targeted delivery of a therapeutic compound utilizing nano-particles to cancerous tumour cells with the multicellular simulator PhysiCell.

Motivation & Objective

• To develop a new evolutionary computation framework that mimics the natural haploid-diploid cycle observed in eukaryotes to improve search efficiency.
• To investigate how the Baldwin effect—phenotypic plasticity influencing evolutionary trajectory—can be embedded in a computational algorithm through diploid representation.
• To evaluate the proposed HDEA method on both abstract fitness landscapes (NK model) and a complex biological simulator (PhysiCell) for therapeutic design.
• To determine whether diploid representation with recombination enhances optimization performance compared to standard haploid evolutionary algorithms.
• To identify optimal parameters for nanoparticle delivery to cancer cells using the HDEA in a multicellular simulation environment.

Proposed method

• The HDEA uses diploid individuals, each composed of two haploid genomes, to represent solutions in the search space, enabling a generalization over two points in the fitness landscape.
• Fitness evaluation combines contributions from both haploid genomes, simulating gene interaction effects such as co-dominance or partial dominance, rather than applying dominance rules.
• Recombination via meiosis is applied during reproduction, allowing genetic mixing between parental genomes and generating novel diploid offspring with hybrid traits.
• The algorithm alternates between diploid selection and haploid gamete formation, mimicking the natural eukaryotic life cycle and enabling phenotypic plasticity through diploid expression.
• The method is tested on the tunable NK model to assess performance across varying fitness landscape ruggedness, and applied to the PhysiCell multicellular simulator for cancer therapy optimization.
• Statistical comparisons are made between HDEA and standard haploid evolutionary algorithms using non-parametric tests (Wilcoxon signed-rank) on 30 independent runs.

Experimental results

Research questions

• RQ1 How does the HDEA method perform compared to standard haploid evolutionary algorithms on abstract fitness landscapes with varying ruggedness?
• RQ2 Can the diploid representation and recombination in HDEA effectively exploit the Baldwin effect to improve search performance on complex, rugged fitness landscapes?
• RQ3 Does HDEA outperform standard evolutionary algorithms in optimizing nanoparticle delivery to cancer cells within the PhysiCell multicellular simulator?
• RQ4 What are the key parameter configurations that lead to optimal therapeutic outcomes in the HDEA-optimized nanoparticle designs?
• RQ5 Is the performance gain of HDEA statistically significant in terms of average and best solution quality across multiple runs?

Key findings

• HDEA achieved significantly better average fitness values than the standard haploid evolutionary algorithm in the NK model (p = 0.0256, Wilcoxon signed-rank test).
• Although the best solutions were not statistically significantly better in the NK model (p = 0.3763), the HDEA found significantly better solutions in the first ten runs (p = 0.0215).
• In the PhysiCell simulation, HDEA reached fitter solutions faster despite high stochasticity in the fitness landscape, indicating robustness and efficiency.
• The cargo release O2 threshold parameter was the most influential, with optimal solutions clustering near 11 mmHg, consistent with prior studies.
• HDEA solutions showed a skew toward lower worker relative repulsion and higher worker motility persistence time, suggesting a distinct parameter space preference.
• Boxplots and scatter plots revealed that HDEA produced more consistent and concentrated parameter distributions, especially for key therapeutic design parameters.