July Highlights: Global and Domestic AI Breakthroughs in Biomedicine

➛ Google AI Drug Development Reaches Key Milestone: Anticancer Drug to Begin First Human Trial

On July 7, The STAR Market Daily reported that Colin Murdoch, President of Isomorphic Labs and Chief Business Officer of Google DeepMind, revealed,Isomorphic Labs' AI-Designed Drug Ready to Launch First Human Trial.The company uses AI technology to assist in the development of anti-cancer drugs, "The next major milestone will be when the drug enters clinical trials and is soon to be tested on humans." Isomorphic Labs was spun off from Google's DeepMind in 2021, originating from DeepMind's renowned breakthrough research, AlphaFold.

AlphaFold, a model developed by DeepMind, is an AI tool used for predicting the three-dimensional structure of proteins. The third-generation model was updated last year.AlphaFold 3, known as"Revolutionary model" that can predict the structure and interactions of all biomolecules with unprecedented accuracy.Last year, Isomorphic Labs signed significant research collaboration agreements with Novartis and Eli Lilly, supporting pharmaceutical companies' existing drug programs while also designing internal candidate drugs in fields such as oncology and immunology, which are licensed out after early trials.

For pharmaceutical companies, the investment in new drug development is enormous, with only a 10% success rate in drug trials. However, Isomorphic Labs' technology has the potential to significantly increase this success rate.At present, AI technology is widely applied in the pharmaceutical field, and industrial cooperation is gaining momentum.For example, XtalPi signed a letter of intent with DoveTree LLC to utilize an AI drug discovery platform for the discovery and development of small molecule and antibody candidate drugs targeting tumors, autoimmune diseases, and neurological disorders. In June, AstraZeneca collaborated with CSPC Pharmaceutical Group based on AI-driven drug development, advancing the discovery and development of novel oral candidate drugs for high-priority targets. Data shows that in the first quarter of 2025, at least 14 AI companies received funding from multinational pharmaceutical corporations (MNCs) or initiated collaborations with them. AI analyzes vast amounts of biomedical data to identify potential drug targets, assist in molecular drug design, predict compound structures, activities, and toxicities using machine learning algorithms, and rapidly screen clinical candidate drugs with potential, shortening R&D time. In recent years, multinational pharmaceutical companies have accelerated integration with AI-driven pharmaceutical enterprises through various models. According to Wanlian Securities, pharmaceutical companies paying high upfront fees indicate increased credibility in AI-generated molecules, and collaborative risk-sharing is driving AI-driven drug development from "concept" to "cash flow."

➛ 2025 World Artificial Intelligence Conference: Release of AI Virtual Disease Biologist "MetaLife," Solving the Challenge of Drug Target Discovery

Reported by Oriental Network on July 28, in drug development, target discovery is crucial. However, over 90% of candidate drugs fail in clinical trials globally, with nearly half of the failures traceable to defects in target selection. At the "New Models and Opportunities Ecosystem Forum for AI-Driven Drug Development" during the 2025 World Artificial Intelligence Conference,Lingang Laboratory Releases AI Virtual Disease Biology "OriGene," Sets Multiple Records, Solves Drug Target Discovery Challenges.

Before the widespread adoption of AI, target discovery relied on experienced disease biologists to review literature, analyze omics data, and validate biological hypotheses through multiple rounds of experiments. However, "MetaLife" can overcome the high complexity of target discovery by coordinating multiple specialized agents to reason across multimodal data such as multi-omics information, clinical data, and evidence from patents and publications, enabling systematic and large-scale identification of novel therapeutic targets. Designed with the mindset and analytical methods of professional disease biologists at its core, "MetaLife" "thinks, works, and grows" like a disease biologist, continuously absorbing feedback from human experts and wet-lab experiments through a self-evolving framework to iteratively optimize core analytical workflows and toolkits, improving both accuracy and efficiency.

The R&D team collaborated with Zhongshan Hospital to test the ability of "Isomorphic Labs" to propose original targets for liver cancer and colorectal cancer. In liver cancer treatment research, the "Isomorphic Labs" system identified G-protein-coupled receptor GPR160 as a key target and independently completed the full-chain validation:Clinical data analysis shows that GPR160 expression is elevated in hepatocellular carcinoma tissues compared to normal liver tissues, and high expression correlates with a significant reduction in recurrence-free survival rates. Experimental validation demonstrates that its small-molecule inhibitor potently suppresses the Huh-7 cell line. Results also reveal that blocking this target activates T-cell immune responses, unveiling a "direct killing + immunomodulation" dual mechanism, potentially representing a novel immune checkpoint. Significant tumor-suppressive effects were observed in tumor fragment models derived from 3 patients and organoids derived from 12 patients, paving a new path for precision treatment of liver cancer.

Targeting the bottleneck in colorectal cancer treatment, "Isomorphic Labs" intelligently selects the competitive target arginase ARG2. Clinical evidence shows that ARG2 is highly expressed specifically in cancer tissues, with a system-designed autonomous validation loop:The first round confirmed the dose-dependent inhibition of the inhibitor in the HCT116 cell line, followed by an independent upgrade of the experimental protocol based on feedback, achieving potent tumor suppression responses in PDO models from four metastatic patients. This success clears a key obstacle for targeted therapy translation in colorectal cancer. The "Yuansheng" system has demonstrated the ability to discover original targets under the guidance of scientists and iterative experimental feedback. Moving forward, the R&D team will expand collaborations and strengthen its core capabilities, aiming to develop it into a "virtual disease biologist" that plays a pivotal role in drug discovery, continuously delivering high-value targets and injecting a "target pipeline" into new drug development.

➛ AI Revolutionizes Antibody Discovery with Groundbreaking Progress: Chai-2 Model Shows Remarkable Results

Recently, ChaiDiscovery announced that its AI model, Chai-2, has achieved revolutionary progress in the field of antibody discovery. According to the 21st Century Business Herald, a scientist used this AI system to solve an antibody design problem, which previously cost $5 million, within hours, whereas traditional research methods would take months or even years.Latest data shows that Chai-2 successfully discovered antibodies with a 16% hit rate on an experimental board costing only 10 yuan, demonstrating its potential and value in antibody development and bringing new possibilities to the biopharmaceutical industry.

Chai-2 is a multimodal generative AI model developed by Chai Discovery, focusing on molecular structure prediction and design. Unlike traditional antibody discovery methods (animal immunization or high-throughput screening), it does not require existing antibody templates or large-scale experimental screening. It can design the complementarity-determining regions (CDRs) of antibodies from scratch using only target antigen and epitope information. In tests on 52 novel antigen targets, Chai-2 achieved a success rate of 16%-20% with only 20 designs tested, and 50% of the targets yielded at least one effective binder—far surpassing the 0.1% success rate of traditional AI methods. Industry analysts believe that Chai-2’s breakthrough represents a milestone for AI in drug discovery, signaling a shift toward the engineering transformation of biology. In the future, as Chai-2 continues to optimize in areas such as manufacturability and pharmacokinetics, AI-driven drug discovery could achieve the goal of "one-shot design success," bringing revolutionary advances to fields like cancer, autoimmune diseases, and infectious diseases.

➛ AI Empowers Neuroscience Research, Cognition Therapeutics Makes Progress in Clinical Trials

Cognition Therapeutics Announces Significant Progress in Neurological Disease Clinical Trials, Particularly in Alzheimer’s Disease and Dry Age-Related Macular Degeneration (AMD) Research. Enrollment in the Phase II START Study for Early Alzheimer’s Disease Exceeds 50%, Aiming to Include up to 540 Individuals. The Trial is Supported by an $81 Million Grant from the National Institute on Aging and Simultaneously Evaluates the Potential of Zervimesine in Reducing Lesion Growth in Dry AMD Treatment. Reports from the Phase II MAGNIFY Trial for Geographic Atrophy Secondary to Dry AMD Show that Zervimesine Slowed Lesion Growth by 28.6% and Reduced Lesion Size by 28.2% Compared to Placebo.

Cognition Therapeutics CEO Lisa Ricciardi pointed out that in the research of diseases such as Alzheimer's, AI technology can help identify and analyze biomarkers, enabling precision medicine approaches. Neurological disorders were once a research "black hole," and breakthroughs in AI technology are expected to bring new hope.

➛ AI Protein Design Platform Upgrade Accelerates Development of Biopharmaceutical and Biomanufacturing Industries

At the 2025 World Artificial Intelligence Conference, according to a report by Cover News on July 27,AI protein design company Molecular Heart showcased the fully upgraded AI protein optimization and design platform MoleculeOS, along with dozens of solutions for the biopharmaceutical and biomanufacturing industries.MoleculeOS integrates more than ten AI protein prediction, optimization, and design technologies, including the multimodal AI protein foundational model NewOrigin (Darwin), combined with scientific computational methods such as molecular dynamics and quantum chemistry. In protein design, Molecular Heart achieves ultra-high precision molecular dynamic structure prediction and protein dynamic design, with molecular simulation accuracy reaching industrial-grade levels. MoleculeOS encapsulates various AI algorithms into automated workflows, refining drug design, enzyme design workflows, and solutions, which have been validated in multiple industry projects. It can meet real industry application needs such as innovative drug design and synthetic biology with "one-click" customization of proteins with specific functions. Additionally, MoleculeOS features a conversational AI agent, enabling biologists without an AI background to quickly and accurately design high-value molecules through dialogue with the AI.

In the past, scientists designed proteins through experimental screening in laboratories, which was time-consuming, labor-intensive, and had a low success rate. Nowadays, with the empowerment of AI, these industry pain points are being gradually resolved. The new method of "AI design + minimal experimental validation" in generative biology significantly reduces the operational burden on biologists, enhances R&D efficiency, and the high-value molecules designed by AI also improve the success rate of research and development.

➛ AI Computing Power Boosts Fudan Medical Research Breakthrough; Alzheimer's Early Screening and Diagnosis Kit to Launch Within the Year

Shanghai, July 19th, by reporter from China News Service: Fudan University has made a breakthrough achievement in the medical field,The early screening and diagnosis reagent for Alzheimer's disease (AD) will be available in major hospitals and体检 centers by the end of this year.A novel therapeutic target for Parkinson's disease (PD) was previously discovered. These studies were supported by AI computing power from the CFFF platform, jointly developed by Fudan University and Alibaba Cloud, among others.

Professor Yujin Cai from Huashan Hospital, Fudan University, stated that biopharmaceutical big data requires robust algorithms, computing power, and new algorithm support. In 2023, China's largest cloud-based scientific research intelligent computing platform, CFFF, went online, utilizing Alibaba Cloud technology to form a "supercomputer." Among this, the Alibaba Cloud Ulanqab Data Center provides over a thousand parallel intelligent computing cards for research projects, supporting the training of large models with hundreds of billions of parameters. Professor Yujin Cai’s team relied on the CFFF platform to predict the risk of Alzheimer’s disease 15 years in advance, with an accuracy exceeding 98.7%. The findings were published in the journal *Nature*. Additionally, they discovered a new therapeutic target for Parkinson’s disease and used AI to screen candidate drugs. The research achievements were featured in top international journals such as *Cell* and *Nature*.

Alzheimer's disease and Parkinson's disease are neurodegenerative disorders that pose a serious threat to human health, making early warning and precise intervention crucial. Traditional research methods handle limited data, consume significant time, and are inefficient. After the establishment of the CFFF platform, researchers replaced the traditional "hypothesis-driven" model with a "data + algorithm" approach, enabling the processing of more data in less time. In the field of Alzheimer's disease, Yu Jintai’s team tested over 6,361 cerebrospinal fluid proteomics datasets, using AI computing power to screen out five key proteins, increasing diagnostic accuracy to 98.7%. In the field of Parkinson's disease, with the help of AI computing power and large model technology, the team was able to screen potential targets across all genes, predict protein structures, and virtually screen small-molecule compounds, completing in five years what would have taken decades. Currently, Alibaba Cloud's AI infrastructure supports the comprehensive upgrade of the CFFF platform, providing open access to multidisciplinary models and scientific datasets, facilitating the publication of multiple high-level papers.

This article is for academic sharing only. Please indicate the source for reprinting. If there is any infringement, please contact WeChat: bioonSir for deletion or modification!