Prompteina
The first all-atom protein structure generation model that supports natural-language specification of design objectives. It enables de novo protein design through text conditioning and supports residue-level alignment fine-tuning for fine-grained structural control.
Timeline (last 90 days)
Prompteina is the first all-atom structure generation model that supports specifying design goals in natural language.
Prompteina works analogously to text-to-image generation: a natural language description is provided to generate the corresponding three-dimensional protein structure.
Prompteina is adapted from an existing structure generation model, Proteina-Complexa, rather than being trained from scratch.
Prompteina's training pipeline consists of two steps: text-conditioned pretraining and residue-level alignment fine-tuning.
Prompteina outperforms the baseline model on 15 out of 18 structural properties.
In binder design, Prompteina leads the baseline on 13 out of at least 20 targets and improves interface confidence by 14% compared to the strongest baseline.
After incorporating iterative in silico feedback, on the hardest targets, Prompteina's gain over non-iterative generation is 57% higher than that of the second-best baseline.
The research team identified 9 binders against GM2A, with the strongest apparent dissociation constant reaching 107 nM.
All Facts (8)
Prompteina is the first all-atom structure generation model that supports specifying design goals in natural language.
50%UnverifiedPrompteina works analogously to text-to-image generation: a natural language description is provided to generate the corresponding three-dimensional protein structure.
50%UnverifiedPrompteina is adapted from an existing structure generation model, Proteina-Complexa, rather than being trained from scratch.
50%UnverifiedPrompteina's training pipeline consists of two steps: text-conditioned pretraining and residue-level alignment fine-tuning.
50%UnverifiedPrompteina outperforms the baseline model on 15 out of 18 structural properties.
50%UnverifiedIn binder design, Prompteina leads the baseline on 13 out of at least 20 targets and improves interface confidence by 14% compared to the strongest baseline.
50%UnverifiedAfter incorporating iterative in silico feedback, on the hardest targets, Prompteina's gain over non-iterative generation is 57% higher than that of the second-best baseline.
50%UnverifiedThe research team identified 9 binders against GM2A, with the strongest apparent dissociation constant reaching 107 nM.
50%