On November 19, Baitu Biotech officially released the AI ​​drug safety evaluation system (hereinafter referred to as the "AI Safety Evaluation System"), which combines virtual cell models, toxicity prediction capabilities with non-clinical safety evaluation services, and relies on safety prediction models such as "Virtual Monkey" to extend AI predictions from the cellular level to the animal body.

The official release of the AI ​​safety assessment system also marks that the company's virtual cell capabilities have begun to move from public evaluation to the real drug development process, and have further extended from cell state prediction to empowering animal safety evaluation in vivo, helping the R&D team to determine which candidate molecules are worth prioritizing and which risks need to be verified before investing in animal experiments.

Currently, Baitu Biotech has established cooperation with more than 30 CROs, experimental institutions and other partners to provide integrated services for drug research and development covering exploratory toxicology, application toxicology, toxicokinetics, histopathology and application support.

At the press conference, Zhang Xiaoming, senior vice president of Baitu Biotechnology and head of AI R&D, introduced the AI ​​drug safety evaluation system for the first time.

From virtual cells to in vivo prediction in animals, allowing key judgments to occur earlier

In recent years, artificial intelligence has been widely used in target discovery and molecular design, significantly expanding the scope of exploration in the early stage of drug research and development. However, when candidate molecules enter the animal experiment stage, problems such as long development cycles, high costs, and slow feedback of results still exist.

Taking experimental monkeys as an example, the purchase price for a single cynomolgus monkey has reached as high as 300,000 yuan, and the dosing phase of some long-term toxicity tests takes about six months. For the large number of molecules to be screened in early stage development, the cost and time of conducting animal trials one by one is prohibitive. The U.S. Food and Drug Administration (FDA) is also promoting the use of new methods (NAMs) such as computational models for drug safety evaluation and reducing animal testing in the development of some antibodies.

Especially in the safety evaluation process, from dose design, dosing observation, to pathological film reading, data review and report issuance, all the work is closely linked. If some risks are not discovered until late in the trial, they may lead to supplementary trials, data rework, and even adjustments to the R&D plan. Baitu Biotech has launched an AI safety assessment system, hoping to make key judgments in advance before deciding which molecules will enter animal testing. In this process, virtual cells can predict the likely responses of candidate molecules, providing early toxicity clues.

In 2025, Baitu Biotech xTrimoSCPerturb defeated thousands of teams from 114 countries around the world and won the championship in the world's first virtual cell challenge, verifying the model's ability to predict cell state changes under unknown disturbances. On this basis, Baitu Biotech further combined existing animal test data to explore virtual drug delivery through safety prediction models such as "virtual monkeys". Based on this, pharmaceutical companies can prioritize candidate molecules with more potential, invest limited trial resources into the molecules that need to be verified most, reduce later rework and the time and costs it brings, and gain time for subsequent development and commercialization.

After entering the actual drug administration stage, the cooperative CRO verifies the risk prediction suggested by the model through animal experiments. Baitu Biotech matches the prediction results of the same molecule with real experimental data for verification and iteration of the model, and precipitates safety evaluation data that can be used for subsequent research and development. With the advancement of data accumulation and model verification, Baitu Biotech hopes to gradually expand its toxicity prediction capabilities covering cells, organs and animals, corroborate prediction results at different levels with real test data, provide pharmaceutical companies with a more comprehensive safety evidence chain, and gradually reduce their reliance on animal testing.

The AI ​​safety assessment system released this time does not provide a single toxicity prediction result, but embeds AI capabilities into key nodes of non-clinical safety evaluation, forming a continuous process of model prediction, experimental execution, data analysis and report delivery.

Before the start of the trial, the model can integrate in vitro cell toxicology, small animal in vivo data, and historical results of large animals and similar molecules to provide a reference for species selection, dose grouping, safety window and first toxic organ judgment, and help the R&D team identify some risks before starting formal animal experiments.

During the trial, the model can continuously analyze in-process data such as body weight, food intake, clinical testing, and toxicokinetics, identify abnormal trends, deviations from the relationship between dose and exposure, and potential data gaps, and assist the research team in timely verification and handling of problems.

After entering the anatomy and reporting stage, the model can also complete pathological pre-screening and suspicious area annotation, mark areas that require focus, and conduct consistency checks on original data, report forms, and application materials, allowing pathologists and researchers to devote more time to professional judgment.

By forwarding some analysis and verification work, the AI ​​safety assessment system hopes to reduce waiting, serial processing and late rework in the process, allowing toxic signals to be identified earlier than formal reports, allowing reports to enter the review and issuance process earlier, and allowing key judgments to occur earlier.

Achievements in multiple benchmarks, with models identifying drug toxicity earlier

One of the core models supporting the AI ​​safety assessment system is the xTrimoTox toxicity prediction model built by Baitu Biotech based on the Cell World model.

Focusing on key safety evaluation tasks such as hepatotoxicity, cardiotoxicity, genotoxicity and acute toxicity, Baitu Biotech conducted a systematic evaluation of xTrimoTox on the TDC public benchmark. The results show that in the test tasks selected this time, xTrimoTox achieved higher results than the public leading model. Among them, the AUROC of the model reached 0.9637 and 0.8918 on drug-induced liver injury and hERG cardiotoxicity tasks respectively, and reached 0.9076 in the blind test of liver injury of 47 clinical drugs. AUROC is a commonly used indicator to measure the model's ability to distinguish between high-risk and low-risk samples. The closer the value is to 1, the stronger the model's discriminating ability. This shows that the predictive ability of xTrimoTox is not limited to a single toxicity endpoint, but can integrate molecular structure and cellular response information to maintain relatively stable performance in different types of safety evaluation tasks.

Baitushengke further independently tested the model on the Tox21 original challenge data set. Tox21 is a representative public evaluation system in the field of toxicity prediction. It mainly examines whether compounds will affect nuclear receptors and cellular stress pathways. Among Tox21's 645 official test molecules and 12 toxicity endpoint evaluations, the average ROC-AUC of xTrimoTox reached 0.8598, which is higher than the classic champion model DeepTox (0.8470), as well as public leading models such as CLAMP+RF (0.8440), SNN (0.8440), and CheMeleon (0.8380).

Join forces with CROs and industry experts to build an AI safety assessment collaboration system

When AI enters the safety evaluation process, it requires not only model capabilities, but also a mature experimental system, professional talents, and project delivery capabilities.

Focusing on the AI ​​safety assessment system, Baitu Biotech has established a cooperation network with more than 30 domestic CROs and experimental institutions with professional execution capabilities. Partners will provide support in non-clinical experiments, toxicokinetics, bioanalysis and histopathology based on the trial types, technical requirements and application goals of different projects. Baitu Biotechnology is responsible for AI models, solution support, institutional matching, process tracking and technical review, jointly connecting predictions, experiments, data analysis and application delivery.

This cooperation model not only retains the CRO organization's professional capabilities in experimental execution and quality systems, but also enables Baitu Biotech's AI model to enter the real experimental process. The risk tips given by the model can be verified through CRO experiments, and the data generated by the experiments can feed back to the model to promote continuous iteration of toxicity prediction capabilities.

On this basis, Baitu Biotech has also gradually introduced senior consultants and industry experts in the field of safety evaluation to participate in the construction of technical systems, research and judgment of complex projects, and review of model results. Currently, Baitu Biotechnology has cooperated with China Academy of Information and Communications Technology and other institutions to jointly build a life science AI evaluation standard system focusing on practical application scenarios such as AI drug safety evaluation. By bringing together the capabilities of the AI ​​R&D team, CRO partners and industry experts, Baitu Biotech hopes to form an AI safety assessment collaboration system that combines model innovation, experimental verification and professional judgment. At present, relevant capabilities have entered actual project operations.

Zhang Xiaoming, senior vice president and head of AI R&D at Baitu Biotechnology, said: "After winning the first Virtual Cell Challenge last year, we have been thinking about whether the model's prediction of cell response can be used as a basis for judgment by pharmaceutical companies in real research and development. In the past, when we talked about the dry and wet closed loop of AI drug research and development, the industry paid more attention to molecular design and in vitro verification. We hope Through the AI safety assessment system, we can take this closed loop one step further, from molecules to cells and organs, and then to animal safety evaluation, combining cell and organ level predictions with in vivo data, and exploring the use of 'virtual monkeys' to carry out virtual drug delivery. The experimental results will in turn test and improve the model, bringing new assistance to drug research and development."

Liu Tengfei, Vice Minister of the Future Industry Department of the Industrial Internet and Internet of Things Research Institute of the China Academy of Information and Communications Technology, said: "Whether AI safety assessment can exert value depends on whether model prediction, experimental verification and comprehensive evaluation can form a reviewable chain of evidence. Baitu Biotech has the world's leading virtual cell model, and also has the ability to rely on its self-built high-throughput experimental platform to carry out dry and wet sealing Environmental verification capabilities. China Academy of Information and Communications Technology will leverage its accumulation in standard development, testing and evaluation, and industrial collaboration to jointly build life science AI evaluation capabilities around data quality, model performance, and credible verification, so as to promote the reproducibility and credibility of AI-assisted safety evaluation results and better integration into processes such as drug research and development and biomanufacturing.”

In the future, Baitu Biotechnology will continue to expand cooperation with CRO institutions, scientific research teams and industry experts, integrate data at different scales such as cells, organs, and animals, explore the application of basic life science models in toxicity mechanism research, organ response prediction, and in vivo safety evaluation, and gradually establish a safety evaluation system that can be experimentally verified and can continue to feed back experiments. Baitu Biotech hopes to use this as a starting point to allow AI to cover more drug development links from molecular design, cell response to in vivo evaluation, providing earlier and more reliable information for candidate molecule screening, risk judgment and R&D decision-making.

Leifeng.com (public account: Leifeng.com)

operations

Resend Email Modify Email