Protein and genome design technologies have witnessed remarkable development with the emergence of new tools that allow users to formulate their requests using natural language instead of relying entirely on complex mathematical and spatial specifications. This trend represents a qualitative leap in how researchers interact with bioengineering software, similar to what is known as natural language programming in computer science.
In this context, an international research team including researchers from China has developed a new tool known as Pinal, which aims to simplify the protein design process. The tool relies on two main stages; the first is called T2struct, which translates text commands into a linear schematic diagram based on the spatial probabilities of amino acids, while the second stage, represented by the SAProt-T system, re-evaluates and approximates the structural framework to produce the final required protein sequence.
Practical Experiments with Limited but Promising Success Rates
To test the efficiency of the Pinal tool, researchers asked it to design four different types of proteins, including a fluorescent protein and multiple enzymes. In the case of fluorescent proteins similar to the famous GFP protein, the tool was directed to generate fifty thousand candidates, which subsequently underwent rigorous filtering processes that compared target structures and configurations.
The filtering process yielded a short list of twenty-nine candidates that were produced and tested in the laboratory, resulting in one functional and fluorescent protein in the desired form. Despite the low direct success rate, researchers point out that this approach reduces the need for deep technical expertise in protein chemistry, provided that laboratories equipped for screening and verification are available.
Expanding the Scope of Natural Language to Include DNA
In a parallel development, Hugging Face, in cooperation with the Zhongguancun Academy and the University of Naples Federico II, announced an open-weight model called Carbon X. This model operates locally and can be run on low-cost hardware such as Apple laptops, solving the privacy dilemmas and cloud computing costs associated with major commercial models.
The model is based on the Qwen3-14B human language model and was retrained using DNA data to become capable of understanding and translating between human languages and genetic codes, enabling users to ask direct analytical questions about the impact of genetic modifications.
What Changes Practically for Users and Researchers?
These developments do not mean dispensing with biologists or specialized laboratories, as there remains an urgent need for advanced screening software and laboratory experimental verification. However, on the other hand, they lower barriers for non-specialist researchers in bioinformatics, facilitating the posing of complex research questions through simple language formulations and accelerating the pace of discoveries in the post-genomic era.