图书情报工作

图书情报工作 >

2026 , Vol. 70 >Issue 15: 91 - 103

DOI: https://doi.org/10.13266/j.issn.0252-3116.2026.15.007

研究论文

5-dBTFM-LLM驱动的生物医学转化阶段可解释识别研究

  • 张云秋 ,
  • 黄麒霏
展开
  • 吉林大学公共卫生学院医学信息学系 长春 130021
张云秋,教授,博士,博士生导师,E-mail:yunqiu@jlu.edu.cn;黄麒霏,博士研究生。

收稿日期: 2025-05-12

  修回日期: 2025-08-07

  网络出版日期: 2026-07-30

收起

5-dBTFM-LLM-Driven Interpretable Recognition of Biomedical Translation Stages

  • Zhang Yunqiu ,
  • Huang Qifei
Expand
  • The Department of Medical Informatics, School of Public Health, Jilin University, Changchun 130021
Zhang Yunqiu, Professor, PhD, Doctoral Supervisor, E-mail:yunqiu@jlu.edu.cn; Huang Qifei, Doctoral Candidate.

Received date: 2025-05-12

  Revised date: 2025-08-07

  Online published: 2026-07-30

Fold

摘要

[目的/意义] 研究提出多维特征与大语言模型驱动的生物医学转化阶段识别方法，旨在提升识别的准确性和可解释性，以辅助生物医学成果的转化评估及层级评价。[方法/过程] 首先，基于转化科学谱系中基础—临床—临床后的转化框架构建涵盖研究对象、研究技术、研究主题、研究领域和临床阶段的多维转化特征框架；然后，基于大语言问答模型的提示学习策略从文献数据中抽取转化特征，并采用SBERT语义嵌入模型实现特征向量化表征；最后，将量化的特征作为输入，构建CatBoost-SHAP可解释集成学习模型，通过特征重要性分析量化各维度特征对转化阶段分类结果的影响程度，实现阶段识别的可解释性。[结果/结论] 实证研究表明，本方法识别的AUC值相比于另外两种主流方法有着显著提升，基于相同数据集取得最优的效果。基础T0阶段分别提升4.0%和2.8%；临床T1/T2阶段分别提升9.1%和4.7%；临床后T3/T4阶段提升3.9%和3.2%。同时，通过特征消融实验也证明五维转化特征对于方法性能提升的有效性。另外，通过可解释性分析还表明特征的重要性排序依次为：研究主题、研究领域、研究技术、研究对象和临床阶段。本研究不仅实现多阶段识别，而且在特征和模型两个方面上提高识别的效果与可解释性，具有显著有效性和一定优越性，可以为后续的科研管理和政策制定等应用场景提供重要的方法支撑。

关键词： 转化研究; 转化科学谱系; AIGC; 集成学习; 可解释性

本文引用格式

张云秋 , 黄麒霏 . 5-dBTFM-LLM驱动的生物医学转化阶段可解释识别研究[J]. 图书情报工作, 2026 , 70(15) : 91 -103 . DOI: 10.13266/j.issn.0252-3116.2026.15.007

Abstract

[Purpose/Significance] This study investigates and proposes an interpretable method for identifying the transformation stages of biomedical research that integrates multi-dimensional features with large language models (LLMs), aiming to enhance the accuracy and interpretability of identification, thereby supporting the transformation evaluation and hierarchical assessment of biomedical research outcomes. [Method/Process] Firstly, a multi-dimensional transformation feature framework covering research objects, research technologies, research topics, research fields and clinical stages is constructed based on the basic-clinical-post-clinical transformation framework in the translational science spectrum. Then, the transformation features are extracted from the literature data based on the prompt learning strategy of the large language question-answering model, and the SBERT semantic embedding model is used to realize the vectorized representation of features. Finally, the quantized features are taken as input to construct a CatBoost-SHAP interpretable integrated learning model. Through feature importance analysis, the influence degree of each dimension feature on the transformation stage classification result is quantified, so as to realize the interpretability of stage identification. [Result/Conclusion] Empirical studies show that the AUC value identified by this method is significantly improved compared with the other two mainstream methods, achieving the optimal effect based on the same data set. The basic T0 stage is increased by 4.0% and 2.8% respectively; the clinical T1/T2 stage is increased by 9.1% and 4.7% respectively; the post-clinical T3/T4 stage is increased by 3.9% and 3.2% respectively. At the same time, the feature ablation experiment also proves the effectiveness of the five-dimensional transformation features in improving the method performance. In addition, the interpretability analysis also shows that the order of feature importance is: research topic, research field, research technology, research object and clinical stage. The study not only realizes multistage identification, but also improves the effect and interpretability of identification in terms of features and models. It has significant effectiveness and certain advantages, and can provide important method support for subsequent application scenarios such as scientific research management and policy formulation.

Key words： translational research; translational science spectrum; AIGC; ensemble learning; explainability

参考文献

[1] 詹启敏, 杜建. 论医学科技与“ 国之重器”[J]. 北京大学学报（医学版）, 2022, 54(5): 785-790. (ZHAN Q M, DU J. On medical science and technology and “national strategic pillars” [J]. Journal of Peking University (health sciences), 2022, 54(5): 785-790.)
[2] 孙蔚, 杨宏钧, 陈中. 建立新型的研究型医院，加速和提高转化医学及临床实验的研究水平[J]. 生物医学, 2014, 4(2): 7-16. (SUN W, YANG H J, CHEN Z. Establish research hospital setting to accelerate translational research and standardize clinical trails [J]. Hans journal of biomedicine, 2014, 4(2): 7-16.)
[3] AUSTIN C P. Translating translation[J]. Nature reviews. adrug discovery, 2018, 17(7): 455-456.
[4] ALVING B, DAI K, CHAN S H H. Translational medicine - what, why and how: an international perspective[M]. Basel: S.Karger AG, 2012: 126.
[5] 陈竺. 推动转化医学发展，应对人民健康挑战[J]. 中国科技奖励, 2011(3): 6-7. (CHEN Z. Promoting translational medicine development to address public health challenges [J]. China science and technology awards, 2011(3): 6-7.)
[6] SURKIS A, HOGLE J A, DIAZGRANADOS D, et al. Classifying publications from the clinical and translational science award program along the translational research spectrum: a machine learning approach[J]. Journal of translational medicine, 2016, 14(1): 235.
[7] 刘江峰, 张冉, 王东波, 等. 追溯科学融合的历史轨迹： AIGC赋能交叉科学测度研究[J]. 图书情报工作, 2024, 68(7): 137-150. (LIU J F, ZHANG R, WANG D B, et al. Tracing the historical trajectory of scientific integration: AIGC empowers interdisciplinarity measurement research [J]. Library and information service, 2024, 68(7): 137-150.)
[8] CESARIO A, GALETTA D, RUSSO P, et al. The role of the surgeon in translational research[J]. The lancet, 2003, 362(9389): 1082.
[9] SUNG N S, CROWLEY W F, GENEL M, et al. Central challenges facing the national clinical research enterprise[J]. Journal of the American Medical Association, 2003, 289(10): 1278-1287.
[10] DEVIEUX J G. Cultural adaptation in translational research: field experiences[J]. Journal of urban health: bulletin of the New York Academy of Medicine, 2005, 82(2 Suppl 3): iii82-91.
[11] BIRMINGHAM K. What is translational research[J]. Nature medicine, 2002, 8(7): 647-647.
[12] WESTFALL J M, MOLD J, FAGNAN L. Practice - based research - “blue Highways” on the NIH roadmap[J]. Journal of the American Medical Association, 2007, 297(4): 403-406.
[13] DOUGHERTY D, CONWAY P H. The “3T’s” road map to transform US health care - the “how” of high - quality care[J]. Journal of the American Medical Association, 2008, 299(19): 2319-2321.
[14] National Center for Advancing Translational Sciences (NCATS). Translational science spectrum[EB/OL]. (2025-08-16) [2025- 08-16]. https://ncats.nih.gov/about/about-translational-science/ spectrum.
[15] 来茂德. 转化医学：从理论到实践[J]. 浙江大学学报（医学版）, 2008(5): 429-431. (LAI M D. Translational medicine: from theory to practice [J]. Journal of Zhejiang University (medical sciences), 2008(5): 429-431.)
[16] 桂永浩. 转化医学：用多学科交叉策略推动医学发展[J]. 复旦 教育 论坛, 2007(6): 86-87, 91. (GUI Y H. Translational medicine: advancing medical science by stimulating interdisciplinary research [J]. Fudan education forum, 2007(6): 86-87, 91.)
[17] HAN X, WILLIAMS S R, ZUCKERMAN B L. A snapshot of translational research funded by the national institutes of health (NIH): a case study using behavioral and social science research awards and clinical and translational science awards funded publications[J]. PloS one, 2018, 13(5): e0196545.
[18] MOLLOY E J, BEARER C F. Translational research is all - encompassing and lets everyone be a researcher[J]. Pediatric research, 2021, 90(1): 2-3.
[19] WEBER G M. Identifying translational science within the triangle of biomedicine[J]. Journal of translational medicine, 2013, 11(1): 126.
[20] KE Q. Identifying translational science through embeddings of controlled vocabularies[J]. Journal of the American Medical Informatics Association, 2019, 26(6): 516-523.
[21] LI X, TANG X, LU W. Tracking biomedical articles along the translational continuum: a measure based on biomedical knowledge representation[J]. Scientometrics, 2023, 128(2): 1295-1319.
[22] 陶斯宇, 谭宗颖. 生物医学领域基础研究转化的测度方法研究 —— 以干细胞和免疫学领域国家自然科学基金资助面上项目产出论文为例[J]. 图书情报工作, 2023, 67(14): 50- 61. (TAO S Y, TAN Z Y. Research on measurement method for translation of basic biomedical research: taking the papers from general projects funded by the national natural science foundation of China in the field of stem cell and immunology as examples [J]. Library and information service, 2023, 67(14): 50-61.)
[23] HUTCHINS B I, DAVIS M T, MESEROLL R A, et al. Predicting translational progress in biomedical research[J]. PLOS biology, 2019, 17(10): e3000416.
[24] KIM Y H, LEVINE A D, NEHL E J, et al. A bibliometric measure of translational science[J]. Scientometrics, 2020, 125(3): 2349-2382.
[25] MAJOR V, SURKIS A, APHINYANAPHONGS Y. Utility of general and specific word embeddings for classifying translational stages of research[J]. AMIA annual symposium proceedings, 2018: 1405-1414.
[26] HSU Y Y, CLYNE M, WEI C H, et al. Using deep learning to identify translational research in genomic medicine beyond bench to bedside[J]. Database, 2019: baz010.
[27] LI X, TANG X. Characterizing interdisciplinarity in drug research: a translational science perspective[J]. Journal of informetrics, 2021, 15(4): 101216.
[28] 吴娜, 刘畅, 刘江峰, 等. AIGC驱动古籍自动摘要研究： 从自然语言理解到生成[J]. 图书馆论坛, 2024,44(9):111- 123. (WU N, LIU C, LIU J F, et al. AIGC-driven research on automatic summarization of ancient classics: from natural language understanding to natural language generation [J]. Library forum, 2024, 44(9): 111-123.)
[29] 张吉玉, 张均胜, 乔晓东. 辅助新颖性评估的科技论文评述画像构建方法[J]. 情报理论与实践, 2023, 46(1): 159-167. (ZHANG J Y, ZHANG J S, QIAO X D. Constructing review profile of scientific article for novelty evaluation assistance [J]. Information studies: theory & application, 2023, 46(1): 159-167.)
[30] 黄少年, 王焕然, 李佩霖. 基于大语言模型的群体性突发事件事理图谱构建与演化分析[J]. 情报杂志, 2024, 43(12): 161-169. (HUANG S N, WANG H R, LI P L. Construction and analysis of the event evolution graph for mass emergency based on the large language model [J]. Journal of intelligence, 2024, 43(12): 161-169.)
[31] 杨冬菊, 黄俊涛. 基于大语言模型的中文科技文献标注方法[J]. 计算机工程, 2024, 50(9): 113-120. (YANG D J, HUANG J T. A Chinese scientific literature annotation method based on large language model [J]. Computer engineering, 2024, 50(9):113-120.)
[32] 张凯, 吕璐成, 韩涛, 等. “ 论文— 专利” 关联视角下的新兴技术识别研究[J]. 情报理论与实践, 2024, 47(9): 183-191. (ZHANG K, LYU L C, HAN T, et al. Research on emerging technology identification from the perspective of “paper - patent” correlation [J]. Information studies: theory & application, 2024, 47(9): 183-191.)
[33] PROKHORENKOVA L, GUSEV G, VOROBEV A, et al. CatBoost: unbiased boosting with categorical features[C]// BENGIO S, WALLACH H, LAROCHELLE H, et al. Advances in neural information processing systems 31. La jolla: Neural Information Processing Systems, 2018.
[34] 付姣, 梁丽芝. 融合权利要求语义特征的专利价值早期预测研究[J]. 情报杂志, 2024, 43(2): 183-191. (FU J, LIANG L Z. On early prediction of patent value by fusing semantic features of claims [J]. Journal of intelligence, 2024, 43(2): 183-191.)
[35] 殷敏, 李晓辉, 李常宝, 等. 一种法律判决预测的影响因素分析方法[J]. 计算机与现代化, 2021(4): 37-41. (YIN M, LI X H, LI C B, et al. An analysis method of influencing factors of legal judgment prediction [J]. Computer and modernization, 2021(4): 37-41.)
[36] LUNDBERG S M, LEE S I. A unified approach to interpreting model predictions[C]//GUYON I, LUXBURG U V, BENGIO S, et al. Advances in neural information processing systems 30(NIPS 2017). La Jolla, USA: Neural Information Processing Systems, 2017.
[37] BORDONS M, ZULUETA M A. Comparison of research team activity in two biomedical fields[J]. Scientometrics, 1997, 40(3): 423-436.
Options
文章导航

/

〈 〉