by Roberto
Fig. 1. Evo, a genomic foundation model. Source
Two years ago I played, feeling simultaneously amused and bemused, with ChatGPT. It was clear artificial intelligence (AI) had taken a quantum leap. While I did not get to play with it, reading about Evo last week evoked similar feelings; it left me both in awe and slightly unnerved. In case you missed it, Evo does with genome sequences what ChatGPT does with words, i.e. almost everything.
Here at STC we do not usually post on the most recent developments on matters microbial. I'm making a bit of an exception here, not to tell you about the details reported in the paper describing Evo. Mainly, because I can lay no claim to really understanding the inner workings of this model. Rather, I feel compelled to share my largely emotional reactions to the ultrafast rate of innovation in biological research afforded by AI as made evident by Evo.
Before going there, it's important to describe Evo. To do that, I feel better simply quoting the legend to the figure above: " Evo, a 7-billion-parameter genomic foundation model, learns biological complexity from individual nucleotides to whole genomes. Trained on 2.7 million raw prokaryotic and phage genome sequences, Evo is naturally multimodal, enabling the codesign of DNA, RNA, and protein molecules that form higher-order functional systems. Evo is also inherently multiscale, enabling prediction and generation tasks at the level of molecules, systems, and genomes." After training, the model predicted the effect of single mutations on an organism's fitness, generated new synthetic CRISPR-Cas molecular complexes, new transposable systems and even predicted megabase-long sequences displaying "plausible genomic architecture." Wow, double wow!
I admit that reading the paper was way over my head. Similarly, the author's summary, the expert commentary and even the "newsy" piece, proved challenging. In the end, I could not decide if this accomplishment was another step in the expected path of computational biology or a quantum leap. My hunch goes with the latter. Now to my reactions.
On the one hand, I cannot but feel utterly amazed by the power of AI-based "large language models." If Evo represents the first baby step in using very large sequence datasets to train such models, where will be in ten years' time? To me this acceleration of progress is extremely exciting, and I think it holds great promise. Not only as a tool for engineering microbes but also for gaining new understanding of the fundamentals of life.
On the other hand, as the authors point out, there are important ethical issues that arise that urge all scientists involved to proceed with caution. That's one issue. Then there is the personal aspect. I was left feeling that the pace is accelerating so fast that I am out of the game, in wonder but also wondering if I have any role left to play. Perhaps that, more than anything else, it why I found Evo slightly unnerving. Still, I was excited enough reading about it that I'll be happy to cheer along from the sidelines, even if at times I feel I no longer fully understand the game.
Do you want to comment on this post? We would be happy about it! Please comment on Mastodon or Bluesky.
Comments