Biotech Industry · global
How Are AI-Designed Proteins Validated? Sino Biological Launches High-Throughput Experimental Services
4RISE integrates protein production, binding analysis, and functional testing. The company says some workflows can deliver binding data in as little as three days; this can accelerate candidate screening, but multiple hurdles remain before efficacy and safety can be demonstrated.
Computers can design protein sequences, but predictions alone cannot answer whether a molecule can be produced, whether it actually binds its target, or what effects it produces after binding. On November 13, Sino Biological (義翹神州) announced the launch of 4RISE, a high-throughput wet-lab validation service that brings AI-designed protein binders into the laboratory for testing, aiming to shorten the wait between design and experimental results.
According to the company’s announcement, 4RISE pairs the XPressMAX cell-free protein synthesis platform with binding analysis. The company claims that the protein expression step can be completed in approximately three hours and that applicable workflows can go from sequence to binding data suitable for computational analysis in as little as three days. These two timelines cover different tasks and should not be interpreted as meaning that all candidates can undergo comprehensive validation within three days.
The service offers four workflows based on the stage of research: initial screening identifies miniproteins or single-domain antibodies with binding signals from unpurified supernatants, followed by standardized comparisons using purified candidates. When more detailed assessment is needed, affinity and association and dissociation kinetics can be measured. Specialized molecular formats can undergo customized mammalian cell expression and validation, providing corresponding tools for initial screening and subsequent confirmation.
Binding measurements use Carterra and Biacore surface plasmon resonance (SPR) platforms and the Octet biolayer interferometry (BLI) platform to analyze interactions between candidates and their targets. The company says structured results can be accessed in real time through an application programming interface (API). For teams that repeatedly refine sequences, this feedback can help guide the next round of design and candidate ranking; whether the data can improve models still depends on experimental quality and how the data are used.
The company’s service page also presents a case involving more than 300 AI-designed miniprotein binder variants: the expression success rate was approximately 50% to 60% when using HEK293 cells, rising to more than 99% after switching to a cell-free system. This is a specific case reported by the supplier, measuring whether proteins could be expressed. It does not mean that the same proportion of molecules can bind their targets effectively, and it cannot be treated as a drug development success rate.
To extend assessment beyond binding, the company lists more than 20 developability assays, more than 1,000 protein bioactivity assays, and more than 10 cellular functional assays. Stability, aggregation propensity, and effects in cells are all additional pieces of information needed when selecting candidates. These are optional services, and the actual testing must still be determined by project requirements.
This announcement concerns a research service, rather than evidence of a new therapy’s efficacy. The available announcement does not provide large-scale independent validation sufficient to comprehensively compare different AI models, targets, and molecular formats, nor does it include human trial results. Faster access to binding data can help eliminate unsuitable designs earlier, but candidates intended for therapeutic use must still progressively establish evidence of function, in vivo effects, and safety.