Replace the slow macs/msprime coalescent simulators with other “generators” (like GAN)

[gregorgorjanc]

Should we continue to face slow coalescent simulations of founder genomes (due to large Ne, etc.) we could supply some pre-simulated genomes or we could use other “generators”. Here is one such "generator", where Generative Adversarial Network (GAN) is trained on real data and can then generate real-like data.

Latent generative modeling of long genetic sequences with GANs
https://www.biorxiv.org/content/10.1101/2024.08.07.607012v3