Phage homing endonuclease amplifies anti-defense genes to evade bacterial immunity - Nature Communications Phages employ diverse counter-defense strategies to overcome bacterial immune systems. Here, the authors reveal that the phage homing nuclease SegB facilitates immune evasion by promoting the segmenta...

Happy to share our new work on how phages escape bacterial immunity.

We show that a phage homing endonuclease drives segmental amplification of anti-defense genes, pointing to a versatile and rapid mode of adaptation.

www.nature.com/articles/s41...

#PhageSky #MicroSky

I am so excited to share our new findings with you! We provide the structural evidence for a direct protein-to-DNA information pathway, showing how a bacterial enzyme 'reads' its own structure to 'write' DNA. www.science.org/doi/10.1126/...

Very proud of this work and all the efforts from my team and collaborators on this! You can now use DGRs for in vivo targeted hypermutagenesis in E. coli. We also included some early proof of concept in Yeast thanks to @seth-shipman.bsky.social !

Excited to share our new findings in @science.org on how the DRT3 bacterial defense system uses a reverse transcriptase that builds DNA repeats without a nucleic acid template. Microbes never cease to amaze!

www.science.org/doi/10.1126/...

Thanks!!

Thanks!!!

Machine learning models by the Bernheim and Laub labs vastly expand the predicted space of bacterial immunity, demonstrating that we are still at the beginning of our exploration 🔎

Congrats @mdmlab.bsky.social and @pdeweirdt.bsky.social et al!

How diverse is bacterial immunity ?

We report in @science.org how language models allowed us to predict 2.4M antiphage proteins spanning >23K novel potential systems.
👏 @emordret.bsky.social, @alexhv.bsky.social & al doi.org/10.1126/scie...

Explore them here defensefinder.mdmlab.fr/wiki/refseq_...

Excited to see our work out in Science today! Using machine learning to identify prokaryotic immune systems www.science.org/doi/10.1126/...

Thanks Jose!

Thanks Luuk!

Seems like great work and will be soooo useful! Congrats :)

A big advance in the field of bacterial immunity, from the Bernheim and Laub labs

Thanks Rotem :) !

2.39 million antiphage proteins in 32,000 bacterial genomes!! Congratulations Aude & team!

Thanks Owen!

Thanks Ilya!!

Nice highlight of both Laub's lab and our work !
www.science.org/doi/10.1126/...

Still so much to discover in the amazing world of bacteria immunity 🧬🦠🛡️

‘Treasure trove’ of antiviral proteins could inspire powerful molecular tools Two research teams mined genomic data from bacteria to create databases containing thousands of antiviral defence proteins that could inspire powerful biotechnologies.

Bacteria have been fighting off viruses using a huge arsenal of molecular weaponry that scientists did not know about — until now. Researchers have identified proteins that could lead to virus-fighting drugs and technologies.

go.nature.com/4dqQnXI

We are more than ever convinced that bacterial immune systems remain one of the richest reservoirs to explore to discover novel biology, biotechnologies and components of immunity shared across domains of life.

We’re excited to see complementary work from @pdeweirdt.bsky.social from Laub's lab. Their approach also uses machine learning to identify prokaryotic immune systems! Complementary approaches, same message: bacterial immunity is much broader than we appreciated.
doi.org/10.1126/scie...

To explore this crazy diversity, we’re incredibly excited to share all these predictions with the community! Not just as huge tables, but also through an interactive web atlas defensefinder.mdmlab.fr/wiki/refseq_....

Search by sequence, PFAM, and more — and have fun hunting for new mechanisms.

Bacterial antiphage systems often function as multi-gene operons.

To make this diversity easier to explore , we developed a method to automatically infer candidate operons for experimental follow-up. We call them PAPO (Predicted AntiPhage Operons).

At pangenome scale, the models predict 2.39 million antiphage proteins across >32,000 bacterial genomes.

More than 85% of predicted defense-associated protein families had no prior link to immunity.

In other words: we are still only exploring a fraction of bacterial immune diversity.

Yes!! We developed GeneCLR, a model that integrates both context and function. It performed better than all of our other models, reaching 99% precision and 92% recall for prediction of antiphage proteins.

So what does GeneCLR-DF tell us about bacterial immunity ?

Then, we fined tuned ESM-DF, a protein language model classifier. Here the underlying signal would be homology (capturing domains and very distant variants). We validated its predictions in E. coli uncovering six additional systems.

But could we get the best of both worlds? Context AND sequence ?

We first developed ALBERT-DF, a model trained on genomic context, hypothesizing it would recognize defense islands or MGE. We validated its predictions in Streptomyces Albus, uncovering six novel antiphage systems, including only proteins with completely novel domains.

While super diverse, known defense systems share recurring features in both protein sequence and genome organization, (ex. specific domains or enrichment in defense islands).

💻 We reasoned that language models trained on protein sequences and genomic context could learn these signals !

🦠 Bacteria encode a highly diverse repertoire of
antiviral systems, with more than 250 systems already experimentally validated, encompassing a broad range of molecular mechanisms.

Rate of discovery: around 1 system per week.

So, are we close to having discovered them all ?