Biotech R&D · global
Getting AI-Designed Proteins into Experimental Testing Faster: Linking Cell-Free Synthesis with Real-Time Binding Analysis
After synthesizing 200 nanobodies, the research team screened them in 4.5 hours to identify 11 candidates that bind to their targets. The workflow shortens the wait for early validation, but many hurdles remain before drug efficacy can be demonstrated.
Computers can design protein structures, but laboratories still need to confirm whether those proteins can be produced and bind to their intended targets. As the number of AI-designed candidates grows, this validation process has become a research and development bottleneck. An integrated workflow announced by Sino Biological and Cytiva attempts to link protein synthesis directly with binding analysis, reducing the wait for cell culture and individual sample purification.
An article published by Drug Target Review on September 11 describes how the team tested this approach using AI-designed VHHs, also known as nanobodies. According to publicly available information from Sino Biological, 200 VHHs were synthesized in parallel within three hours. The reaction supernatants were then used directly for screening, identifying 11 positive binders within 4.5 hours. These times cover synthesis and screening separately and cannot be treated directly as the total time from AI design to complete validation.
The first part of the workflow uses cell-free protein synthesis (CFPS): the cellular machinery responsible for producing proteins is extracted and combined with nucleic acid templates, amino acids, and energy-supplying components to carry out reactions in test tubes. The kit used in this case was CFKIT02. Without having to wait for living cells to grow, researchers could produce multiple candidates simultaneously.
The second part uses Biacore 8 series instruments to track molecular binding and dissociation through surface plasmon resonance (SPR). Using His tags, researchers capture candidate proteins from unpurified supernatants onto a sensor surface, then measure their interactions with the target. “Purification-free” means eliminating the separate purification step for each sample; the analysis itself still relies on selective surface capture.
To check whether crude samples affected the readings, the team compared the affinities of three candidates under four preparation conditions: CFPS supernatants generated using linear and plasmid templates, purified CFPS proteins, and purified proteins produced by Chinese hamster ovary (CHO) cells. The vendor described the results as highly consistent. The publicly available values support similar binding performance for these few molecules, but are insufficient to conclude that all proteins would yield the same results.
Key information remains undisclosed in this case: the public description does not reveal the design model, training data, specific targets, or thresholds for determining positive results for this batch of candidates. The article published in Drug Target Review also carries a Sino Biological byline and describes the same case as the company materials, so it cannot be considered an independent replication. The 11 positive results therefore describe only this screening exercise and cannot represent the general success rate of AI protein design.
For drug discovery, the practical use of this approach is to obtain evidence of candidate binding more quickly, helping researchers decide which designs warrant further investment. Whether these candidates can alter cellular function, remain stable in the body, and be safe still requires separate validation. Accelerating the return of experimental feedback to the design process is the better-supported value of this workflow at present.