Sun Yat-sen University to Transfer Three AI-Powered Patents for Antibody Binding Site Prediction at RMB 80,000

Recently, Sun Yat-sen University issued a public notice regarding its proposed transfer of three invention patents to Guangdong MAGIGENE Technology Co., Ltd. This transaction covers“A Deep Learning Method for Predicting Binding Sites on Antibodies via Sequence Prediction,” “A Protein Function Prediction Method Combining Multi-Task Learning and Self-Attention Mechanisms,” and “A Method for Predicting Protein-Protein Interaction Sites Based on Deep Graph Convolutional Networks”Core patents, proposed transfer amount is80,000 yuan。

This patent focuses on AI-driven technologies for predicting protein functions and key sites, with its primary inventors being a professor at the School of Computer Science and Engineering, Sun Yat-sen University, and the chief engineer of the National Supercomputer Center in Guangzhou.Yang Yuedonget al. Professor Yang Yuedong’s research interests include multi-omics big data mining, end-to-end intelligent drug design, and the development of high-performance computing platforms for biomedicine.

In the fields of oncology, autoimmune diseases, and infectious diseases, antibody-based therapeutics have become a cornerstone of precision medicine. Taking cancer immunotherapy as an example, monoclonal antibodies targeting PD-1/PD-L1 significantly improve survival outcomes for patients with melanoma, non-small cell lung cancer, and other malignancies by blocking immunosuppressive signals and reactivating T-cell-mediated tumor cell killing. In autoimmune conditions such as rheumatoid arthritis, anti-TNF-α antibodies have also become first-line treatment regimens.

However, the current development of antibody drugs still heavily relies on the traditional experiment-driven model. The entire process, from target discovery and antibody screening to affinity optimization, is characterized by long cycles and high costs, while the approval rate for candidate molecules remains low. The contradiction between efficiency and cost is becoming increasingly prominent.

The limitations of current clinical protocols are primarily manifested at three levels:

First, the efficiency of target discovery is low.Traditional protein functional annotation relies on wet-lab validation. Elucidating protein–protein interaction sites requires complex techniques such as cryo-electron microscopy and X-ray crystallography, which are not only technically challenging but also constrained by sample quality and equipment availability, thereby preventing the rapid identification of numerous potential drug targets.

Second, the antibody development cycle is lengthy.Screening for mutations in the antigen-binding sites (CDR regions) of antibodies relies on high-throughput experiments such as phage display and yeast display, making it difficult to accurately predict the impact of mutations on affinity. This often necessitates multiple rounds of iterative optimization, thereby delaying drug market entry.

Third, inadequate response to public health emergencies,During outbreaks of emerging infectious diseases, the key to rapidly developing neutralizing antibodies lies in elucidating the interaction interface between viral proteins and host receptors; however, the timeliness of traditional experimental techniques fails to meet the demands of emergency response, resulting in significant delays in clinical treatment and drug development.

The urgent clinical need to shorten R&D cycles and reduce the risk of failure is driving a transformation in R&D models from “experiment-driven” to “computation-driven,” which constitutes the core background for the emergence of the technologies in this patent portfolio.

The three core technologies in this patent portfolio have achieved multidimensional breakthroughs in the field of AI-assisted drug development.

InAlgorithm Architecture, its antibody binding site prediction technologyFirst Integration of Transformer Bidirectional Encoding with BiLSTM Network, enabling precise identification of key antigen-binding residues solely from antibody amino acid sequences without relying on three-dimensional structural information, thereby overcoming the limitations of traditional methods that depend on crystal structures; a multi-task learning framework leverages self-attention mechanisms to simultaneously predict multiple functional attributes from individual protein sequences, significantly enhancing the efficiency of protein function annotation and eliminating the redundant training costs associated with traditional single-task models; modeling techniques based on deep graph convolutional networks construct graph models of protein three-dimensional structures, accurately capturing spatial features at protein-protein interaction interfaces, with prediction accuracy surpassing that of traditional machine learning methods.

AtR&D EfficiencyIn terms of this aspect, the technology significantly shortens the time required from antibody sequence input to binding site output, markedly improving efficiency compared to traditional experimental methods. Meanwhile, it reduces reliance on costly experiments such as cryo-electron microscopy and high-throughput screening, effectively lowering the R&D cost per project.

Furthermore, the technology features cross-scenario adaptability, making it applicable not only to antibody drug development but also extendable to scenarios such as vaccine target design, structural elucidation of protein complexes, and research on virus-host interaction mechanisms. It supports both cloud computing and lightweight local deployment models, enabling seamless integration with existing R&D workflows and lowering the adoption barrier for small and medium-sized pharmaceutical enterprises and research institutions.

The Value of This AI-Assisted Drug Discovery Patent Portfolio Aligns Deeply with the Global Biopharmaceutical Industry“Cost Reduction and Efficiency Enhancement, Source Innovation”core demands, boasting both vast market potential and diverse commercialization opportunities.

From a market outlook perspective, the global AI-driven drug discovery industry is in a phase of rapid development. As a core segment within this field, antibody drug discovery is experiencing sustained growth in demand, providing ample application scenarios for this technology. In terms of commercial value, the technology features clear and diversified revenue models: it can provide technical services such as target prediction and antibody screening to pharmaceutical companies and CROs, charging on a per-project basis or taking a share of the outcomes; it can establish joint R&D platforms with pharmaceutical companies to develop candidate drugs in hot areas such as ADCs and bispecific antibodies, monetizing the technology through license-out arrangements; and it can expand into the scientific research services market by offering protein function research tools to universities and research institutes, thereby covering the entire chain from basic research to preclinical development.

In the long term, the implementation of this technology will drive the transformation of China’s AI-driven drug R&D from “follow-on innovation” to “pioneering innovation,” empowering domestic pharmaceutical companies to seize a competitive edge in the race for emerging targets, while also responding to national"Integration of Industry, Academia, Research, and Application"Guided by policy, it provides efficient tools for the R&D of emergency drugs for infectious diseases and the exploration of treatment regimens for rare diseases. Its technological iterations and data accumulation will establish sustained industry barriers, securing a significant position in the global competitive landscape of AI-driven drug discovery.