The Laboratory of Evolutionary Design has released the Proto language to integrate artificial intelligence models for generative biology. Led by Brian Hie, the research team developed this framework to connect isolated computational tools for experimental biologists. Consequently, the system allows researchers to design biological sequences by specifying high-level functional objectives.
Traditionally, biological research has relied on assembling natural components through trial-and-error methods. However, this new framework composes specialized artificial intelligence models and bioinformatics tools to predict biological structures. Specifically, the system produces sequences that satisfy multiple design objectives simultaneously.
Core Primitives of the Framework
The framework distills biological design into four abstract primitives, which include sequences, generators, constraints, and optimizers. Sequences define molecular strings, while generators propose candidates for testing. Meanwhile, constraints score these candidates, and optimizers steer the generation process toward desired biological functions.
The Proto language defines modularity at a functional level rather than at the sequence level. As a result, the system relies on deep generative models to design de novo sequences that satisfy specified constraints. Users can compile these high-level constraints into a unified energy function that mathematical optimizers minimize.
Experimental Results in Biological Design
Researchers tested the framework on synthetic promoter-repressor pairs to evaluate its performance. Notably, the system achieved functional designs after experimentally testing only tens of candidates. This represents a high success rate for the de novo design of DNA-protein interactions.
Furthermore, the team applied the framework to design cell-line-specific alternative splicing in human cell lines. The system successfully targeted several cell line combinations, including brain versus blood, after testing only tens of designs. In contrast, previous methods required testing thousands of designs to obtain functional results.
Technical Architecture of the Proto language
Computational users can access the Proto language through a Python API for fine-grained control. Additionally, the platform features a drag-and-drop graphical web interface designed for experimental biologists. This dual-interface approach aims to make generative biology tools accessible to a wider scientific audience.
To resolve software conflicts, the framework uses a dependency isolation system. Specifically, each computational tool runs within its own virtual environment through subprocesses dispatched from the main process. This architecture manages hardware resources and parallelization to maximize computational efficiency.
Future Outlook and Open Source Access
The research team has made the framework open source to encourage community contributions. They designed the system to serve as a programming language that both human scientists and AI coding agents can use. Consequently, the developers anticipate rapid integration of newly developed models by the scientific community.
The integration of these tools could transform research in health technology by reducing setup times. “What we want is to spare researchers the sleepless nights we had figuring out dependencies,” said co-author Ben Viggiano. The software is currently available for public access.
Source: arcinstitute.org
