Yunha Hwang

Gaia — Tatta Bio

We are building this infrastructure for the scientific community, and we invite feedback and collaboration from researchers at every stage. We are grateful to
the Moore Foundation for their generous support in making this project possible. Stay tuned for more updates!

www.tatta.bio/gaia

Today's sequence data infrastructure is set up for failure in the age of AI. Building an open and collaborative sequence platform for both Human and AI scientists.

At Tatta Bio, we have been thinking deeply about the sequence-to-function problem. We believe that before AI can power functional prediction, we first need to rethink how we curate, manage, and share sequence data. Here, we share our initial ideas on what we are building next:

Assemblies of long-read metagenomes suffer from diverse errors Genomes from metagenomes have revolutionised our understanding of microbial diversity, ecology, and evolution, propelling advances in basic science, biomedicine, and biotechnology. Assembly algorithms...

I am very happy (and anxious) to share with you our most recent work in which we evaluated four of the most popular long-read assemblers,

www.biorxiv.org/content/10.1...

and tell you just a little bit about it in the following 🧵

I am so grateful for all the support I received from my mentors, colleagues and collaborators over the years: @pgirguis.bsky.social, @sokrypton.org, @simrouxvirus.bsky.social, @alexjprobst.bsky.social, @annedekas.bsky.social

It’s been an incredible journey building Tatta Bio with @ancornman1.bsky.social to advance AI infrastructure for biology, and I will continue to further our mission as chief scientist.

My lab will couple ML and high throughput experimentation to harness the remarkable functional diversity of microbial genomes. If you are excited about the intersection of AI and microbiology, please get in touch!

It’s official! 🎉 I’m thrilled to announce that I will be joining MIT as an assistant professor in a shared appointment between Biology, EECS and Schwarzman College of Computing this fall.

Job Board | Notion Overview

Tatta Bio is growing! We are hiring *two positions* in Business Development and Software Engineering to lead the development of AI-enabled scientific software for open science and biological sequence interpretation. Please check out the job postings at www.tatta.bio/careers and share widely!

Our thoughts too! (stay tuned👀) 😉

As we improve Gaia Agent, we want to hear your feedback on the agent predictions. If you have suggestions on how we can increase its capabilities, please reach out! This was a major collaborative effort with @cong-ml.bsky.social , @joshuakravitz.com @nishantjha.org @ancornman1.bsky.social @Tatta Bio

Gaia Agent: Context-Aware Functional Insights at Scale — Tatta Bio An AI biologist discovers previously uncharacterized systems in the Mtb genome.

We tested Gaia Agent's capabilities with hypothetical genes in Mycobeterium tuberculosis. In our blog, We detail our in silico validation of Gaia Agent-predicted membrane transporter and lanthipeptide biosynthesis loci that were uncharacterized despite decades of Mtb research. Read more:

Like a human biologist, Gaia Agent considers sequence, structure and genomic context to *think* about functions of novel genes, drastically accelerating our ability to predict functions of billions of unannotated proteins across the tree of life.

Can LLM agents discover novel protein functions? Introducing Gaia Agent 🌎 🤖: an AI biologist capable of reasoning across genomic contexts to predict functions of proteins! Gaia Agent is now integrated with Gaia Search at gaia.tatta.bio

If you are at #NeurIPS2024 don't miss @ancornman1.bsky.social's talk on OMG/gLM2 at 9AM! @workshopmlsb.bsky.social East meeting room 11,12

The OMG dataset: An Open MetaGenomic corpus for mixed-modality genomic language modeling Biological language model performance depends heavily on pretraining data quality, diversity, and size. While metagenomic datasets feature enormous biological diversity, their utilization as pretraini...

Excited to be at #NeurIPS this week. @ancornman1.bsky.social will give a spotlight talk at the @workshopmlsb.bsky.social on gLM2/OMG! Please reach out if you want to chat about gLM2/OMG/Gaia and our latest projects😇

www.biorxiv.org/content/10.1...

MIBiG 4.0: advancing biosynthetic gene cluster curation through global collaboration Abstract. Specialized or secondary metabolites are small molecules of biological origin, often showing potent biological activities with applications in ag

Are you working on natural products? We’ve just released version 4.0 of the MIBiG data standard and repository! It now includes 3059 biosynthetic gene clusters, thanks to the combined efforts of 288 expert contributors. A thread: (1/8) academic.oup.com/nar/advance-...

overview of results for PLAID!

1/🧬 Excited to share PLAID, our new approach for co-generating sequence and all-atom protein structures by sampling from the latent space of ESMFold. This requires only sequences during training, which unlocks more data and annotations:

bit.ly/plaid-proteins
🧵

you can search for eukaryotic sequences too, and you might find interesting homology to microbial proteins! (the current database you search against is microbial)

Our Big Fantastic Virus Database (BFVD) is now published NAR! It contains protein structure predictions of major viral clades, enhanced by petabase-scale homology search and it's explorable on the web.
🌐 bfvd.foldseek.com
💾 bfvd.steineggerlab.workers.dev
📄 academic.oup.com/nar/advance-...

Great question, translation tables 11 and 4 should be covered, and we have seen translation table 15 being accounted for in some cases. @apcamargo.bsky.social

and we are live on biorxiv! bsky.app/profile/bior...

Thank you! We are building additional features (e.g. bookmarks, tags, comments), stay tuned for updates!

Great suggestion -- noted!

We cluster all protein embeddings across all 100 retrieved contexts, and then the top 5 most frequently occurring clusters are colored!

This is a fantastic resource! Yes this is possible and we plan on expanding our database in the coming months, it would make sense to include AllTheBacteria

A huge shoutout to nishantjha.bsky.social, Joshua Kravitz, Jacob West-Roberts, apcamargo.bsky.social, simrouxvirus.bsky.social & Andre Cornman for awesome teamwork, & a big TY to those who participated in user interviews. Gaia is in active development so please reach out with ideas or suggestions.

Introducing Gaia: Context-Aware Protein Search Across Genomic Datasets — Tatta Bio Gaia is an embedding-based search engine for sequences.

Blog: www.tatta.bio/blog/gaia
gLM2_embed: huggingface.co/tattabio/gLM...
OG_prot_90: huggingface.co/datasets/tat...

Check out our preprint www.tatta.bio/gaia-paper for benchmarking results. In our manuscript, we showcase how Gaia can be used for annotating uncharacterized #phage proteins and discovering putative biosynthetic gene clusters!

In order to make this search maximally interpretable, we built a web application that integrates existing tools (HMMer, sequence alignments, ESMFold) with genomic context visualizations. Gaia is freely available on gaia.tatta.bio, please share your feedback!

Gaia searches the protein universe comprising 85M clusters across hundreds of thousands of microbial genomes. Embedding-based search takes ~0.2s per sequence, which is at least two orders of magnitude faster than BLASTp, allowing for real-time search.