Illuminating protein space with a programmable generative model.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Proteins exhibit vast diversity, but their full functional potential remains largely untapped.
• Exploring the vast space of possible protein molecules for novel functions is computationally and experimentally challenging.

Purpose of the Study:

• Introduce Chroma, a generative model for designing novel protein structures and sequences.
• Enable conditioning of the generative process for desired protein properties and functions.

Main Methods:

• Utilize a diffusion process respecting polymer ensemble conformational statistics.
• Employ an efficient neural architecture for long-range reasoning in molecular systems.
• Incorporate layers for 3D protein structure synthesis and low-temperature sampling algorithms.

Main Results:

• Chroma successfully designs proteins and protein complexes with desired properties, including symmetry, shape, and semantic constraints.
• Experimental characterization of 310 designed proteins confirmed high expression, proper folding, and favorable biophysical properties.
• Crystal structures of two designed proteins showed excellent atomistic agreement with Chroma's predictions (1.0 Å backbone RMSD).

Conclusions:

• Chroma offers a unified approach to protein design, accelerating the creation of novel protein-based materials and therapeutics.
• This generative model advances the field of protein engineering and synthetic biology.
• Chroma's capabilities extend to incorporating natural language prompts for protein design.