Associate Manager, Drug Discovery
14小时前
Shanghai, 中国
和黄医药
全职
通过电子邮件或 Google 免费
保存此职位并让您的搜索井井有条
创建一个免费帐户来保存职位、创建警报并从仪表板返回此列表。
通过电子邮件或 Google 免费
Position summary
HUTCHMED is seeking an AI Scientist to develop, evaluate and deploy AI and machine learning methods for antibody and ADC discovery. You will combine proprietary experimental data with sequence, structural and conjugate information to improve molecular design, property prediction and candidate selection. Working with computational designers and wet-lab scientists, you will select methods suited to each scientific question and available data, and connect model development to prospective validation and repeated learning cycles.
Key responsibilities
Translate antibody and ADC discovery priorities into modeling tasks, meaningful biological endpoints and measurable evaluation criteria with project teams. Build modeling datasets and informative features with data colleagues, combining sequence, structure, assay and ADC component information while addressing quality, batch effects and label uncertainty. Develop and compare suitable approaches, including statistical and classical machine learning, tree-based models, deep learning, graph neural networks, generative models and multimodal methods. Train, optimize and adapt models for molecular generation and property prediction, including protein language models and other large language models (LLMs) where relevant to the task. Establish rigorous validation using appropriate baselines, sequence-family or campaign holdouts, ablations and prospective experiments; assess uncertainty, generalization and practical value. Work with discovery scientists to select informative experimental batches using active learning, Bayesian optimization or other suitable strategies, balancing multiple molecular objectives. Implement reproducible training, inference and deployment workflows with versioned datasets and models; deliver usable batch tools or APIs using shared computing infrastructure. Use wet-lab feedback to refine datasets and models, investigate failure cases, monitor performance and communicate limitations and validated improvements to discovery teams. Required qualifications MSc or PhD in machine learning, computer science, statistics, computational biology, bioinformatics or a related quantitative discipline; equivalent applied experience will be considered. Typically at least 2 years of hands-on modeling experience, including relevant doctoral research, with evidence of independently developing and evaluating models for scientific or biological problems. Strong Python programming and practical experience with relevant tools such as scikit-learn and PyTorch or comparable frameworks, supported by reproducible data and model workflows. Sound understanding of model selection, statistical evaluation and uncertainty, including limited or noisy data, class imbalance, information leakage and appropriate independent validation. Ability to understand biological measurements, collaborate with experimental scientists and explain how predictions and model limitations affect antibody and ADC decisions. Preferred experience Experience with antibody, protein or ADC property modeling, molecular representations, graph neural networks, generative models or multimodal learning. Experience in one or more of protein language models or LLM training and fine-tuning, active learning, Bayesian optimization, or prospective wet-lab validation.
Key responsibilities
Translate antibody and ADC discovery priorities into modeling tasks, meaningful biological endpoints and measurable evaluation criteria with project teams. Build modeling datasets and informative features with data colleagues, combining sequence, structure, assay and ADC component information while addressing quality, batch effects and label uncertainty. Develop and compare suitable approaches, including statistical and classical machine learning, tree-based models, deep learning, graph neural networks, generative models and multimodal methods. Train, optimize and adapt models for molecular generation and property prediction, including protein language models and other large language models (LLMs) where relevant to the task. Establish rigorous validation using appropriate baselines, sequence-family or campaign holdouts, ablations and prospective experiments; assess uncertainty, generalization and practical value. Work with discovery scientists to select informative experimental batches using active learning, Bayesian optimization or other suitable strategies, balancing multiple molecular objectives. Implement reproducible training, inference and deployment workflows with versioned datasets and models; deliver usable batch tools or APIs using shared computing infrastructure. Use wet-lab feedback to refine datasets and models, investigate failure cases, monitor performance and communicate limitations and validated improvements to discovery teams. Required qualifications MSc or PhD in machine learning, computer science, statistics, computational biology, bioinformatics or a related quantitative discipline; equivalent applied experience will be considered. Typically at least 2 years of hands-on modeling experience, including relevant doctoral research, with evidence of independently developing and evaluating models for scientific or biological problems. Strong Python programming and practical experience with relevant tools such as scikit-learn and PyTorch or comparable frameworks, supported by reproducible data and model workflows. Sound understanding of model selection, statistical evaluation and uncertainty, including limited or noisy data, class imbalance, information leakage and appropriate independent validation. Ability to understand biological measurements, collaborate with experimental scientists and explain how predictions and model limitations affect antibody and ADC decisions. Preferred experience Experience with antibody, protein or ADC property modeling, molecular representations, graph neural networks, generative models or multimodal learning. Experience in one or more of protein language models or LLM training and fine-tuning, active learning, Bayesian optimization, or prospective wet-lab validation.