Anthrogen presents Odyssey: A 102B protein language model of protein substitution according to training with discrete def

Anthrogen presented Odysseyprotein family languages, sequence models and gene expression, protein programming and conditional design. Production models range from 1.2B to 102 parameters. Anthrogen's research group puts Odyssey as a Frontier, a multimodal model of real protein design activities, and notes that the API is at the first end.

What is the problem Odyssey Target?

Protein Design Amino acid sequences pair in 3D structure and functional context. Many previous models received attention, including details in the entire sequence of everyone. Proteins follow geometric constraints, resulting in long-range spatial neighbors in 3D. Anthrogen frames this as a local problem and proposes a new law of diffusion, called coherence, which better fits the domain.

Input representation and tokwen

Odyssey is multimodal. It embeds sequence tokens, structure tokens, and lightweight function calls, then hashes them into a shared representation. For design, Odyssey uses FITITE's Claqur CANCANIZER, FSQ, to convert 3D geometry into transparent tokens. Think of FSQ as an alphabet that allows for an easy learning model like a sequence. Active settings can include domain tags, secondary structure references, orthologous group labels, or short text descriptions. This joint view provides a model for accessing local sequence patterns and long-range geometric relationships

Backbone replacement, compromise instead of self-centeredness

Consistency restores global attention through Iterative, Local Update via Sparse Connection or sequence graph. Each layer encourages the nearest neighbors to agree first, and then propagates that agreement outside the chain and the communication graph. This change is reversed. Attention scales as O(L²) with a good length of L. Anthlogen reports that the consensus scales as O(L) maintaining a long sequence and reporting of a shared Domain. The company also reports significant improvements in selection-level learning at large scales, which reduces brittleness and restarts.

The purpose of training and generation, dicrete variables

Odyssey trains with good difficulty in sequence and plot tokens. The former procedure uses noise masking that corrects the conversion rate. The backward time deviser relearns the sequence and constant coupling and coordination that work together. In essence, the same backward process supports conditional generation and parallel programming. You can grab the scaffold, adjust the motif, mask, add an active marker, and let the model finish resting while keeping the sequence and construction in sync.

Anthlogen reports the corresponding comparisons where language production is performed during the test. The page notes the low intensity of diffenusion training versus the complex, and the low or compensatory training compared to the comparison of the simple combination. In confirmation, deffion models interfere with their masked counterparts, while the masked model tends to fill its rubbing schedule. The company reveals that the models of deficion are distributed in a combination of full proteins, which align with the sequence of the plus plus design.

Key acquisition

1. Odyssey is a family of multein protein models that reduce sequence, structure, and functional context, with production models at 1.2b, 8b, and 102b parameters.
2. Consistency replaces spatial attention in a structure-aware manner that is O(L) and exhibits large-scale learning behavior at scale.
3. FSQ transforms 3D coordinates into pure structural tokens through joint sequencing and structural modeling.
4. Discrete pollution trains the best time denoiser and, in the same comparison, implicit language modeling during testing.
5. Anthlogen reports better performance with 10x less data than competing models, which account for the lack of data in the protein model.

Odyssey is an impressive model because it works in an integrated sequence and structure modeling with FSQ, consistency, and diversity, which enables conditional construction and planning under practical problems. Odyssey scales to 102b parameters with O(L) complexity, which takes the cost of long proteins and improves learning power. Anthlogen reports to clarify the experience expressed in the language by the modeling of joint languages, which corresponds to the goals of composition. The program aims to create a variety of objectives, including potency, specificity, durability and performance. The Research Team emphasizes data efficiency around 10x competitive models, used for domains with specified data.

Look paper, and Technical details. Feel free to take a look at ours GitHub page for tutorials, code and notebooks. Also, feel free to follow us Kind of stubborn and don't forget to join ours 100K + ML Subreddit and sign up Our newsletter. Wait! Do you telegraph? Now you can join us by telegraph.

Michal Sutter is a data scientist with a Master of Science in Data Science from the University of PADOVA. With a strong foundation in statistical analysis, machine learning, and data engineering, Mikhali excels at turning complex data into actionable findings.