2024生成式人工智能GenAI在生物医药大健康行业应用进展报告

Industry Investment Rating - The report highlights that 40% of pharmaceutical executives plan to reinvest cost savings from GenAI into their 2024 budgets, indicating a strong industry focus on GenAI adoption [2] - 60% of companies have set goals to use GenAI for cost reduction or productivity improvement, with 75% considering it a priority for senior management and boards [2] Core Viewpoints - GenAI is poised to revolutionize the biopharmaceutical and healthcare industry by enhancing efficiency and reshaping workflows across drug development, clinical trials, and post-market activities [2] - The integration of GenAI in biopharma is accelerating, with applications ranging from drug discovery to patient education, driven by the need to reduce operational costs and complexity [2] - 2024 is identified as a pivotal year for the large-scale implementation of GenAI in China's biopharma and healthcare sectors [3] GenAI Technology and Applications GenAI Technology Definition and Background - GenAI focuses on generating new, creative data by learning and understanding data distributions, enabling applications in text, image, audio, and video generation [5] - ChatGPT, based on OpenAI's GPT-3.5, exemplifies GenAI's ability to handle diverse text and reasoning tasks, achieving 100 million monthly active users within two months of launch [5] GenAI Application Domains and Cases Multimodal Content Generation - Text generation: ChatGPT and Jasper are used for tasks like automated text generation, Q&A, and summarization, serving companies like Google and Facebook [7] - Image generation: StabilityAI's Stable Diffusion model has attracted over 1 million users globally, enabling AI-generated art through text prompts [8] - Video generation: OpenAI's Sora, launched in 2024, can generate high-quality videos from text inputs, simulating real-world physics and enabling applications in live streaming and remote education [9] Translation - GenAI models outperform traditional machine translation by understanding word coherence and context, producing more natural and accurate translations [11] Content Understanding and Analysis - Tencent Meeting AI Assistant uses GenAI to extract, analyze, and summarize meeting content, improving efficiency in information processing [12] Scientific Research and Innovation - GenAI aids in drug design, material science, and the discovery of new theories and experimental methods in fields like chemistry and biology [13] Key GenAI Technologies Model Training - Model training involves building AI models from scratch, requiring significant data and computational resources, and is suitable for groundbreaking applications like developing new medical diagnostic AI [15][16] Fine-Tuning - Fine-tuning adapts pre-trained models to specific tasks, balancing efficiency and performance enhancement, and is ideal for tasks requiring specialized knowledge, such as medical terminology adaptation [17][18] Retrieval-Augmented Generation (RAG) - RAG combines information retrieval with text generation, enhancing traditional large language models (LLMs) by integrating external knowledge sources, improving answer accuracy and reducing hallucinations [19][20][21] Prompt Engineering - Prompt engineering involves designing effective prompts to guide pre-trained models, offering flexibility and creativity in model outputs without requiring additional training or computational resources [27][28] GenAI Model Development International Models - ChatGPT, developed by OpenAI, has become a global phenomenon with 1.8 billion users by December 2023, driving the development of competing models like Gemini, ErnieBot, and LLaMA [40] - Gemini, developed by Google DeepMind, supports multimodal understanding and generation, with versions optimized for data centers, enterprise applications, and mobile devices [41] - Claude, developed by Anthropic, offers fast and cost-effective language models with extended context windows, enabling tasks like document summarization and complex problem-solving [42] - LLaMA, developed by Meta, is an open-source model series with high efficiency and flexibility, widely adopted in research and development [43][44] Domestic Models - Baidu's ERNIE Bot, known as "China's ChatGPT," excels in Chinese understanding, literary creation, and logical reasoning, with over 33.42 million questions answered on its first day of public release [51] - Alibaba's Tongyi Qianwen, a large-scale language model, has surpassed GPT-3.5 in performance and is accelerating towards GPT-4 levels, with open-source models ranging from 1.8B to 72B parameters [52] - iFlytek's Spark Cognitive Model V3.5 has significantly improved in text generation, language understanding, and logical reasoning, approaching the capabilities of GPT-4 Turbo [53] - Huawei's Pangu Model, a fully self-developed AI model, offers multimodal capabilities and is applied in industries like meteorology, pharmaceuticals, and manufacturing [54] - Tencent's Hunyuan Model, with over 1 trillion parameters, excels in Chinese creation, logical reasoning, and task execution, supporting various industry applications [56] - Zhipu AI's GLM-4 model has achieved performance close to GPT-4, with enhanced multimodal capabilities and faster inference speeds, supporting longer contexts and more concurrency [57] - Baichuan Intelligence's Baichuan3 model, with over 100 billion parameters, outperforms GPT-4 in Chinese tasks and excels in medical evaluations, poetry creation, and logical reasoning [59] GenAI in Biopharma and Healthcare Drug Discovery - GenAI accelerates drug discovery by analyzing multi-omics data, identifying potential targets, and generating drug molecules, significantly reducing the time and cost of traditional methods [64] - Insilico Medicine's PandaOmics platform integrates GenAI for target discovery and validation, achieving a fully automated wet-dry lab cycle in 14 days [65][66] - ShuiMu Molecular's BioMedGPT-10B model integrates literature, molecules, proteins, and knowledge graphs, outperforming specialized models in tasks like molecular property prediction and drug-target affinity prediction [67][68] Molecular Generation Large Molecule Generation - GenAI predicts and designs large molecules like proteins and antibodies, enabling the creation of novel enzymes and accelerating the development of mRNA vaccines and antibody therapies [69] - AstraZeneca's collaboration with Absci uses GenAI to develop improved cancer antibody therapies, while DeepMind's AlphaFold predicts structures of various biomolecules, aiding drug design [70] Small Molecule Generation - GenAI models predict small molecule structures, enabling rapid screening and optimization of lead compounds, as demonstrated by Merck's collaboration with Variational AI and Iktos' AI-driven drug design solutions [72]