伙伴关系
UCB 新增 Cradle 抗体发现项目的AI平台
在 Google 上的首选来源中添加 Unite.AI
摇篮和UCB 宣布是2026年11月4日的战略合作，在选定的UCB抗体发现项目中应用摇篮的AI蛋白设计和工程平台，从命中产生到铅优化和临床前准备。
合作新增 Cradle的端对端系统为UCB扩展的数字抗体发现和工程程序。 该方案将UCB现有的抗体发现平台与计算科学、疾病生物学专门知识以及技术和研究伙伴网络结合起来。 UCB旨在更快地为患者仍然需要更好选择的疾病确定和优化高质量的候选人。
实验结果再培训项目模式
随着UCB科学家使用该平台，实验结果将反馈到针对具体程序的模型中。 公司将由此产生的周期称为复合学习循环，即“实验室内学习”：每一轮实验室证据为下一套设计提供信息。 科学家继续运用其专门知识和判断力，同时优化候选人并评估向潜在候选人发展的进展情况。
Cradle的平台文件解释了基本工作流程。 Assay数据在个人蛋白质的健身景观上训练一个模型。 在下一回合之前，对于早期的命中生成，一个项目可以从一个单一的序列开始，而无需分析数据，也可以从一个吵闹的筛选池开始，例如NGS浓缩计数从显示或选择运动开始。
多产抗体设计
Cradle的抗体文献组将目标优化为绑定性，稳定性，制造性，免疫性以及可开发性。 可测量的目标可以包括特异性和跨反应性，热稳定性和聚合性，表达率，T细胞突起风险，以及去化或氧化等责任。 这具体地说明了公告的要求，即候选人在成功所需的全部财产中得到优化。
用户将目标候选人简介定义为主要目标和制约因素。 该平台Cradle 称平行优化了指定属性，管理权衡，并估计在进行化验前实现每个目标的概率，同时标出不切实际的目标。 一个候选板块也可以在合成和化验支出之前与训练有素的模型进行比对，产生预期的命中率，每个目标的得分分配以及剩余头室的估计。 这些预测支持关于运行、重新设计或停止一轮的决定。
该系统可通过科学家的无码网络应用程序和计算团队的API访问。 研究团队可以检查变种，比较变种，并将自己的设计与摇篮生成的考生放在同一个板块上。 这些外部设计在同样的基础上得分，并成为以后各轮的训练信号，为科学家们自己的假设保留了影响模型的路线。
署长强调科学家领导的发现
UCB首席科学官Alistair Henry称AI是“科学力量增强者”，它应当增强人类在生物学、化学和翻译研究方面的专业知识，而不是为了自身的目的。 He 称将预测模型与湿板验证相结合，有可能帮助设计更好的分子，加速学习，更快地给患者带来高质量的治疗。
当Cradle CEO和联合创始人Stef van Grieken 称 AI超越单个实验或程序，成为科学家跨发现组织工作方式的一部分时，它就会对生物学研发产生最大的影响。 在这个愿景中，UCB科学家获得了一个共同平台，可以在整个公司的生物治疗管道中实施更多的方案、模式和复杂的工程挑战，尽管正式部署涵盖了选定的发现项目。
合作将雄心与重复的实验学习联系在一起：每个实验可以使下一个设计回合更加精锐，而科学家则保留对候选人决定的责任。 对于UCB来说，实际的考验是这些循环是否产生符合约束性、稳定性、可制造性和可开发性要求的抗体，并朝着临床前准备状态迈进。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Partnerships
UCB Adds Cradle’s AI Platform to Antibody Discovery Projects
Add Unite.AI to your preferred sources on Google
Cradle and UCB announced a strategic collaboration on November 4, 2026, to apply Cradle’s AI protein design and engineering platform within selected UCB antibody discovery projects, from hit generation through lead optimization and preclinical readiness.
The collaboration adds Cradle’s end-to-end system to UCB’s expanding digital antibody discovery and engineering program. That program combines UCB’s existing antibody discovery platform with computational science, disease-biology expertise and a network of technology and research partners. UCB aims to identify and optimize high-quality candidates more quickly for diseases where patients still need better options.
Experimental Results Retrain Project Models
As UCB scientists use the platform, experimental results will feed back into program-specific models. The companies describe the resulting cycle as a compounding learning loop, or “lab-in-the-loop”: each round of laboratory evidence informs the next set of designs. Scientists continue to apply their expertise and judgement while optimizing candidates and assessing progression toward a potential candidate.
Cradle’s platform documentation explains the underlying workflow. Assay data train a model on an individual protein’s fitness landscape. Results uploaded from a plate—Cradle’s generations, a team’s rational designs or a mixture—retrain that model before the following round. For early hit generation, a project can start from a single sequence without assay data or from a noisy screening pool such as NGS enrichment counts from a display or selection campaign.
Multi-Property Antibody Design
Cradle’s antibody documentation groups optimization targets into binding, stability, manufacturability, immunogenicity and developability. Measurable objectives can include specificity and cross-reactivity, thermostability and aggregation, expression yield, T-cell epitope risk, and liabilities such as deamidation or oxidation. This gives concrete meaning to the announcement’s requirement that candidates be optimized across the full set of properties needed for success.
Users define a target candidate profile as a primary objective plus constraints. Cradle says the platform optimizes specified properties in parallel, manages trade-offs and estimates the probability of meeting each objective before an assay, while flagging unrealistic targets. A candidate plate can also be scored against the trained model before synthesis and assay spending, producing an expected hit rate, score distributions for each objective and an estimate of remaining headroom. Those forecasts support decisions to run, redesign or stop a round.
The system is accessible through a no-code web application for scientists and an API for computational teams. Research teams can inspect variants, compare mutations and place their own designs alongside Cradle-generated candidates on the same plate. Those external designs are scored on the same basis and become training signal for subsequent rounds, preserving a route for scientists’ own hypotheses to influence the model.
Executives Emphasize Scientist-Led Discovery
UCB Chief Scientific Officer Alistair Henry called AI a “scientific force multiplier” that should augment human expertise across biology, chemistry and translational research rather than be pursued for its own sake. He said combining predictive modeling with wet-lab validation could potentially help design better molecules, accelerate learning and bring high-quality therapeutics to patients sooner.
Cradle CEO and co-founder Stef van Grieken said AI would have its greatest impact on biologics R&D when it extends beyond individual experiments or programs and becomes part of how scientists work across a discovery organization. Within that vision, UCB scientists gain a common platform for more programs, modalities and complex engineering challenges across the company’s biotherapeutics pipeline, although the formal deployment covers selected discovery projects.
The collaboration ties that ambition to repeated experimental learning: each assay can sharpen the next design round, while scientists retain responsibility for candidate decisions. For UCB, the practical test is whether those cycles produce antibodies that meet binding, stability, manufacturability and developability requirements together and progress toward preclinical readiness.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.