[Edaily Reporter KIM JI-WAN ] “It’s because AI itself doesn’t know when it’s wrong.”
Park Woong-yang, CEO of GENINUS Inc. and Director of the Genomics Research Institute at Samsung Seoul Hospital, explained the reason for combining AI-driven new drug development with experimental validation during an interview held on the 9th at the company’s headquarters in Munjeong-dong, Songpa-gu, Seoul. Even targets identified by AI as promising may yield results different from expectations when tested on actual cancer cells. He explained that only by verifying through experiments what is correct and what is incorrect—and then retraining the AI based on those results—can future predictions improve.
Park Woong-yang, CEO of GENINUS Inc. and Director of the Genomics Research Institute at Samsung Seoul Hospital, explains the new drug development strategy based on an “AI feedback loop”—which links AI predictions with experimental validation—at the company’s headquarters in Munjeong-dong, Songpa-gu, Seoul, on the 9th. (Photo: ReporterKIM JI-WAN )
GENINUS Inc.(389030)This is the core of the “AI feedback loop” being promoted by the company. It involves identifying targets using patient data and then feeding the results verified in the laboratory back into the AI. The AI feedback loop is a re-learning structure that combines experimental verification (data × AI × laboratory).
CEO Park stated, “AI trained solely on literature and public data can propose an infinite number of plausible targets, but there is no way to know which of them actually work in real patient cells.” He added, “The only way to find the correct answer is through experimentation, and those experimental results must be fed back into the AI for it to ‘evolve.’”
“Before designing a drug, we determine where to strike.”
GENINUS Inc.’s AI-powered new drug development platform, “Intellibil,” focuses on target discovery—selecting which proteins in cancer cells
to attack
. This means identifying the targets that will determine therapeutic efficacy and safety before designing the drug’s molecular structure.
CEO Park explained, “GENINUS Inc.’s AI isn’t designed to design ‘compounds’; it’s designed to determine ‘where to strike.’”
If the wrong target is chosen, even if the antibody binds effectively, it will be useless. If the target protein is not present in sufficient quantities in the cancer cells, the effect will be limited; if it is also present in normal cells, there is a risk of attacking normal tissue as well.
Intellibil identifies targets by comparing tissue samples from patients who responded to treatment with those who did not. It then adds spatial omics data to determine the location of cells. By distinguishing whether a specific protein is present in cancer cells or in adjacent normal cells, it identifies differences that would otherwise be masked by the average values across the entire tissue.
CEO Park stated, “Intellibil has increased the accuracy of distinguishing between tumor and normal cells from 79% to 98%, reducing errors at the very start of the target selection process,” adding, “The AUC—a predictive performance metric for distinguishing treatment responses—has also improved from 0.756 to 0.924.”
These two metrics demonstrate, respectively, the ability to accurately distinguish which cells to target and the ability to identify patients who will respond to treatment.
CEO Park emphasized, “It’s about selecting targets from the outset that are highly likely to work in Saramin,” adding, “If the target is correct, failures and retries in subsequent stages are reduced, shortening the overall development period.”
Failures Are Also Learning Data… “Validation Results Must Feed Back into the AI”
Once the AI has selected a target, it’s the laboratory’s turn. The lab secures antibodies that bind to that target and verifies whether the expected function is observed in actual cells. GENINUS Inc. is connecting the AI and the laboratory so that these results can be utilized for subsequent predictions.
The process follows this sequence: patient data is generated, the AI predicts targets, the lab validates them, and the results are then fed back into the AI for further learning. Not only successful results but also those that fall short of expectations serve as learning data.
CEO Park pointed out, “Typically, AI companies handle the predictions while pharmaceutical companies handle the validation,” adding, “As a result, data from failed experiments does not make its way back to the AI company.” He continued, “If prediction and validation are separated, the AI may not obtain sufficient information to refine its judgments.”
GENINUS Inc. has eliminated this disconnect by integrating the entire process—from data generation through joint hospital research to AI analysis and antibody validation. This approach involves incorporating questions raised during the analysis into the experiments and reanalyzing issues identified in the experiments.
He emphasized this structural difference, stating, “Because GENINUS Inc. houses both AI and laboratory operations within a single company, we can obtain the necessary answers quickly and efficiently.”
“Even if you start with the same data, the AI will be different a year later.”
CEO Park stated, “The concept of closed-loop, evolving AI models is universal,” adding, “Our differentiation lies not in the concept itself, but in what we use to drive the loop and how quickly we can run it.”
The competitive advantage he highlighted is patient data linked to treatment responses. Simply knowing the characteristics of cancer cells is not enough. He explained that to identify the next point of attack, one must know which drugs were used, who responded to them, and who developed resistance.
CEO Park emphasized, “Data directly generated from patient tissue samples—at the cellular and spatial levels—that includes treatment responses cannot be purchased or downloaded,” adding, “It is the result of ongoing joint research with hospitals since our founding.”
Cancer cells obtained from actual patients are also used for validation. In its in-house laboratories, the company evaluates antibody binding and function using patient-derived cells. This process verifies whether targets identified in patient data are effective in patient-derived cancer cells, and the results are then fed back into the AI for training.
Another competitive advantage lies in the speed at which this process is repeated. No matter how quickly the AI proposes targets, if experiments do not follow, only predictions accumulate. GENINUS Inc.’s strategy is to integrate data production, AI analysis, and experimental validation internally to rapidly secure the evidence needed for subsequent decisions.
He emphasized, “What makes it difficult for competitors to catch up is not the algorithm, but ‘data with verified answers’ and ‘the speed at which we generate those answers,’” adding, “Even if they start with the same data, the AI of a company that runs the loop 10 times will be completely different from that of a company that runs it only twice a year later.”
GENINUS Inc. plans to demonstrate the capabilities of Intellibil through animal testing results. GENINUS Inc. has identified four new breast cancer targets. Currently, the company is securing and screening antibodies for each target while verifying their binding, function, and mechanisms of action.
CEO Park expressed confidence, stating, “Ultimately, the efficacy of these targets must be proven through animal testing and clinical trials,” and added, “The answer to whether the new breast cancer targets identified by Intellibil are effective will be revealed by the results of animal testing in the first half of next year.”
Doubts surrounding LG Energy Solution(373220)’s “AA” credit rating have grown rapidly over the past year. As the company’s debt has ballooned due to continued factory expansions even amid the electric…
“It’s because AI itself doesn’t know when it’s wrong.”Park Woong-yang, CEO of GENINUS Inc. and Director of the Genomics Research Institute at Samsung Seoul Hospital, explained the reason for combining…
DuChemBio Co.,LTD.(176750)continued its revenue growth in the third quarter, driven by increased sales of radiopharmaceuticals for the diagnosis of Alzheimer’s disease. With the domestic launch of Eli…