Accutar Biotechnology Wins DREAM Challenge in Kinase Inhibitor Selectivity Prediction, Accelerating AI-Driven Drug Discovery

Not long ago, DeepMind won the protein 3D structure prediction competition. Although it is still far from having a real impact on pharmaceutical development, it excited the entire biomedical and artificial intelligence academic community. However, the announcement of the results of the DREAM Challenge, a competition focused on predicting the affinity of kinase inhibitors—a class of drugs with direct implications for drug development—has truly energized the biomedical and artificial intelligence industries!

According to the first-round results of the DREAM Challenge announced by the organizers, the Pearson and Spearman correlation coefficients for predicting the affinity of kinase inhibitor drugs have reached their highest levels on record. This indicates that artificial intelligence is transforming the drug development process from blind screening, random trial-and-error, and limited sampling (where experimental screening, no matter how extensive, remains constrained in its sampling capacity) into a more rational and efficient endeavor. The accuracy of target-drug affinity prediction has improved significantly compared with the past, and the promising vision of leveraging AI to accelerate the R&D of kinase inhibitors is gradually becoming a reality.

VCBeat (WeChat ID: vcbeat) learned from the official report that the team winning the championship of the challenge is Accutar Biotechnology, an AI-driven biopharmaceutical company. In this challenge, the Accutar Biotechnology team successfully applied its recently developed 3D conformation prediction and quantitative binding affinity prediction models for drug-target interactions. Based on this approach, they addressed core issues in the pharmaceutical field, such as lead drug discovery, optimization, and improving target selectivity. They achieved first place in all major technical metrics, including Pearson and Spearman correlation coefficients, F1 score (F1-Measure), and mean area under the ROC curve (AUC).

Every cell in the human body is composed of billions of protein molecules, some of which provide structural support while others catalyze biological reactions. Dysfunctional proteins can lead to disease; historically, using drugs to restrict conformational changes in proteins and thereby limit their overactivation has achieved curative effects for various diseases, including multiple types of blood cancers.

Since Gleevec, the first anti-tumor small-molecule inhibitor targeting tyrosine kinases for the treatment of leukemia, achieved widespread acclaim in 1997, dozens of such inhibitors had been approved by 2016. An ideal targeted anti-tumor agent should be characterized by high efficacy, low toxicity, and strong specificity; however, only three of the top ten best-selling targeted anti-tumor drugs in 2017 were small-molecule inhibitors. Although small-molecule inhibitors present substantial market opportunities as targeted cancer therapies, they still face critical bottlenecks that need to be overcome:

I. Small-molecule inhibitors need to compete with the natural ligands of protein kinases.

II. Small-molecule inhibitors may cause off-target toxicity at any time.

III. Small-molecule inhibitors require “24/7 coverage” to ensure that the function of target protein kinases is continuously controlled.

Predicting the binding affinity between protein kinases and small-molecule inhibitors is one of the key applications of artificial intelligence in drug discovery. Understanding the binding affinity between protein kinases and small-molecule inhibitors can rapidly help us gain insights into biological processes, structural biology, and the driving forces behind structure–function relationships. If small molecules with the highest binding affinity to target proteins can be accurately identified from vast compound libraries through computational screening, it can significantly reduce the labor costs and financial expenditures associated with experimental screening.

Given the critical role of protein kinases in the onset and progression of various diseases, this challenge focuses on small-molecule inhibitors of protein kinases. It aims to evaluate the capability of machine learning models to predict the binding affinity between protein kinases and small-molecule inhibitors, assess the accuracy of participating systems in predicting single drug-kinase binding interactions, and place greater emphasis on predicting the target selectivity of drug candidates.

Selectivity for the target is a key benchmark and area of improvement for next-generation small-molecule drugs. Generally, the ability to predict selectivity signifies a major step forward in achieving more precise “targeting” by targeted therapies, shifting drug development closer to “rational design” rather than “trial-and-error,” enhancing specific competitiveness against protein kinases, and reducing off-target toxicity.

The DREAM Challenge aims to address one of the most critical issues in the pharmaceutical industry—predicting the binding affinity between protein kinases and small-molecule inhibitors. Consequently, it is regarded as the most influential algorithm competition in the field of international computational biomedicine. Since 2006, the challenge has been held annually for 12 consecutive editions. In each edition, different organizing committees release their proprietary data and design task-specific modeling challenges for participants. Its third-party validation framework ensures algorithmic reproducibility, thereby enabling the most rigorous and effective validation of predictive models.

The latest official information indicates that more than 300 research teams from around the world participated in this year’s DREAM Challenge on predicting the binding affinity of kinase inhibitor drugs, including teams from the U.S. National Institutes of Health, various academic institutions, and other industrial participants who chose to remain anonymous.

Challenge Champion: Iceland Spar Biotechnology Co., Ltd. (hereinafter referred to as “Iceland Spar”) is a biotech startup leveraging artificial intelligence for new drug development. The company has achieved significant results in areas such as novel drug discovery, drug repurposing, drug screening, and prediction of drug properties, and has established extensive collaborations with numerous major pharmaceutical companies both in China and abroad. Iceland Spar has completed a joint investment round led by IDG Capital and Yitu Technology. Dr. Fan Jie, the CEO, conducted his doctoral studies under Professor Nikola Pavletich, a member of the U.S. National Academy of Sciences, and completed his postdoctoral training under Dr. Gunter Blobel, a Nobel Laureate.

Iceland Spar aims to accelerate preclinical drug development by driving AI-guided global optimization of various drug properties—including potency, target selectivity, and medicinal chemistry and pharmacokinetic indices—thereby reforming the traditional trial-and-error drug development model that often requires dozens of iterative cycles. Iceland Spar has successfully trained 3D conformational prediction models for drug-target binding and quantitative models for predicting binding affinity. Leveraging this approach, the company addresses core challenges in the pharmaceutical industry, such as lead discovery, optimization, and enhancement of target selectivity. By improving screening efficiency and optimizing structure-activity relationships, Iceland Spar is committed to significantly shortening new drug development timelines and substantially reducing costs.

Following the competition, Fan Jie, CEO of Iceland Spar, stated: “The prediction of targets and drug affinity is a critical battleground in the pharmaceutical industry. Iceland Spar’s top performance in this competition has bolstered our confidence and strengthened our resolve to drive this industrial revolution. We hope that the outcomes of this competition will set a benchmark for the emerging field of AI-guided drug development. By addressing the practical needs and pain points of the traditional pharmaceutical industry, we aim to promote the healthy and sustainable development of this sector and strive to genuinely solve real-world problems in pharmaceutical research and development.”