PDF is available at www.biorxiv.org/content/10.1... and I'd be very curious to get takes from @hsalis.bsky.social, @akosnyerges.bsky.social the @cultivarium.bsky.social folks and anybody else!

My conclusion is that this is a cool model that nobody understands how to use yet. Genomes aren’t human-readable, and Evo gives us a hint how we might bridge that gap to start reading, designing and generally operating on the scale of organisms instead of single proteins/operons.

The generation of novel CRISPR systems & genomes doesn't seem very meaningful to me - I'm not convinced that those sequences do anything. In particular the genomes probably aren't even at the level of "colorless green ideas sleep furiously"

The part on zero-shot predictions on proteins & RNAs was neat, but doesn't seem revolutionary. They can match other predictive algorithms for fitness of protein/RNA. There's a ceiling on what these models can currently do and they've run into it by using a lot more data. Sure?

Figure 4E/S8 is exciting. Evo is good at predicting transposon boundaries with entropy, but I bet it encodes more data than that. Why is entropy high at the start of TnpA? What does it look like for bigger operons? Can we use this to explain biological functionality & essentiality at operon-scale?

Yeah, I definitely applaud this effort despite the open-source sacrifice. Having a validated, verified & high-quality genome seq for a repository stock is worth a lot.

It's like when Addgene started doing whole-plasmid sequencing, except it's a lot harder to do organisms than plasmids!