Help us design a £50m experiment to make evolution predictable

We're looking for input from data scientists, geneticists, and evolutionary biologists on how to design a large-scale experiment that would let us predict evolution — not just explain it after the fact.

Evolutionary processes can explain almost any biological outcome after it happens — and predict almost none. If we could change that, the implications would reach across biologics, conservation, crops, pest management, biosafety, and cancer.

As part of ARIA's £50m Engineering Ecosystem Resilience funding, we're scoping an experiment to test how far predictability can be pushed. The rough shape we're considering:

Use a eukaryote with a short generation time (one candidate: the nematode Caenorhabditis remanei).
Run experimental evolution at unprecedented scale, with genomic characterisation throughout.
Train predictive models on the resulting dataset.
Test them through blind prediction competitions, in the spirit of CASP for protein structure.

The open questions we'd love your help with:

Which species should we use?
How many generations, what population sizes, and how many populations?
What selective conditions and starting setups?
What should we measure — sequencing, RNA-seq, phenotyping?
How much data is needed to reach predictive power within budget?
Where can automation take us?

This thinking is being developed with Richard Nichols and Simon Evans, among others. If you work on any piece of this — from experimental evolution to large-scale genomics to predictive modelling — we want to hear how you'd design it.