深度学习在创新小分子药物研发中的应用进展

Advances in application of deep learning in innovative small-molecule drug discovery and development

摘要

随着生物医药与人工智能的深度融合，深度学习在创新小分子药物研发中的应用已由早期的辅助筛选与性质预测，逐步扩展至覆盖“分子设计-分子合成-活性测试”的关键环节，并涌现出从靶点/先导发现到临床验证的代表性实践。相较于主要依赖经验规则、局部数据和离线评分的传统计算机辅助药物设计方法，深度学习能够依托大规模多模态数据，学习可迁移的分子、靶点、反应和表型表征，并通过生成式模型、结构预测、自动化合成和高通量测试平台的协同，实现对化学空间的可控探索、多目标优化和“设计-验证”闭环迭代。在部分代表性研究中，深度学习已显示出提高候选分子富集效率、缩短先导发现周期并降低无效实验验证规模的潜力。本文围绕深度学习赋能创新小分子药物研发的关键环节，重点综述其在分子设计、合成实现与生物活性测试中的应用进展，并结合具有湿实验验证的高影响力研究与进入临床阶段的代表性案例，分析其在真实研发场景中的转化价值。最后，本文进一步讨论当前面临的数据质量与标准化不足、可合成性约束不强、模型评价与研发决策脱节、跨学科闭环集成难度大等挑战，为深度学习驱动的小分子药物创新与转化研究提供参考。

Abstract

With the deep integration of biomedicine and artificial intelligence, the application of deep learning in innovative small-molecule drug discovery and development has gradually expanded from its early use in assisted screening and property prediction to key stages encompassing "molecular design-molecular synthesis-bioactivity testing", and representative applications spanning target and lead discovery to clinical validation have emerged. Compared with conventional computer-aided drug design methods that mainly rely on empirical rules, local datasets, and offline scoring, deep learning can learn transferable representations of molecules, targets, reactions, and phenotypes from large-scale multimodal data, and, through the integration of generative models, structure prediction, automated synthesis, and high-throughput testing platforms, enable controllable exploration of chemical space, multi-objective optimization, and closed-loop "design-validation" iteration. In some representative studies, deep learning has shown the potential to improve the enrichment efficiency of candidate molecules, shorten the lead discovery cycle, and reduce the scale of ineffective experimental validation. This article focuses on the key stages of deep learning-enabled innovative small-molecule drug discovery and development and reviews advances in molecular design, synthetic implementation, and bioactivity testing. By examining high-impact studies with wet-lab validation and representative clinical-stage programs, it further analyzes the translational value of deep learning in real-world drug discovery and development. Finally, this article further discusses current challenges, including insufficient data quality and standardization, inadequate constraints on synthesizability, the disconnect between model evaluation and R&D decision-making, and the difficulty of interdisciplinary closed-loop integration, providing a reference for deep learning-driven small-molecule drug innovation and translational research.

关键词

深度学习 / 药物研发 / 分子设计 / 分子合成 / 活性测试 / 临床转化

Key words

Deep learning / Drug discovery and development / Molecular design / Molecular synthesis / Bioactivity testing / Clinical translation

参考文献

[1] WU K, KWON S H, ZHOU X, et al. Overcoming challenges in small-molecule drug bioavailability: A review of key factors and approaches[J]. Int J Mol Sci, 2024, 25(23): 13121.

[2] MINIKEL E V, PAINTER J L, DONG C C, et al. Refining the impact of genetic evidence on clinical success[J]. Nature, 2024, 629(8012): 624-629.

[3] SINDT F, BRET G, ROGNAN D. On the difficulty to rescore hits from ultralarge docking screens[J]. J Chem Inf Model, 2025, 65(11): 5553-5566.

[4] SABE V T, NTOMBELA T, JHAMBA L A, et al. Current trends in computer aided drug design and a highlight of drugs discovered via computational techniques: A review[J]. Eur J Med Chem, 2021, 224: 113705.

[5] ZHANG K, YANG X, WANG Y, et al. Artificial intelligence in drug development[J]. Nat Med, 2025, 31(1): 45-59.

[6] DURANT G, BOYLES F, BIRCHALL K, et al. The future of machine learning for small-molecule drug discovery will be driven by data[J]. Nat Comput Sci, 2024, 4(10): 735-743.

[7] YANG X, WANG Y, BYRNE R, et al. Concepts of artificial intelligence for computer-assisted drug discovery[J]. Chem Rev, 2019, 119(18): 10520-10594.

[8] LI M, SONG K, HE J, et al. Electron-density-informed effective and reliable de novo molecular design and optimization with ED2Mol[J]. Nat Mach Intell, 2025, 7(8): 1355-1368.

[9] XU Y, CUI H, PANG K, et al. LUMI-lab: A foundation model-driven autonomous platform enabling discovery of ionizable lipid designs for mRNA delivery[J]. Cell, 2026, 189(6): 1620-1635.e25.

[10] JIANG Y, ZHANG G, YOU J, et al. PocketFlow is a data-and-knowledge-driven structure-based molecular generative model[J]. Nat Mach Intell, 2024, 6(3): 326-337.

[11] YU J L, ZHOU C, NING X L, et al. Knowledge-guided diffusion model for 3D ligand-pharmacophore mapping[J]. Nat Commun, 2025, 16(1): 2269.

[12] PENG J, YU J L, YANG Z B, et al. Pharmacophore-oriented 3D molecular generation toward efficient feature-customized drug discovery[J]. Nat Comput Sci, 2025, 5(10): 898-914.

[13] HU Q, SUN C, HE H, et al. Target-aware 3D molecular generation based on guided equivariant diffusion[J]. Nat Commun, 2025, 16(1): 7928.

[14] PENG X, GUO R, GUO F, et al. Unified modeling of 3D molecular generation via atomic interactions with PocketX-Mol[J]. Cell, 2026, 189(7): 1904-1922.e28.

[15] YANG Z, SUN W, HUANG Q, et al. Deep learning-assisted discovery of a potent and cell-active inhibitor of RNA N6-methyladenosine recognition protein YTHDC2[J]. Nat Commun, 2026, 17(1): 46.

[16] SWANSON K, LIU G, CATACUTAN D B, et al. Generative AI for designing and validating easily synthesizable and structurally novel antibiotics[J]. Nat Mach Intell, 2024, 6(3): 338-353.

[17] WANG M, LI S, WANG J, et al. ClickGen: Directed exploration of synthesizable chemical space via modular reactions and reinforcement learning[J]. Nat Commun, 2024, 15(1): 10127.

[18] JIA Y, GAO B, TAN J, et al. Deep contrastive learning enables genome-wide virtual screening[J]. Science, 2026, 391(6781): eads9530.

[19] ZHOU G, RUSNAC D V, PARK H, et al. An artificial intelligence accelerated virtual screening platform for drug discovery[J]. Nat Commun, 2024, 15(1): 7761.

[20] CAO D, CHEN M, ZHANG R, et al. SurfDock is a surface-informed diffusion generative model for reliable and accurate protein-ligand complex prediction[J]. Nat Methods, 2025, 22(2): 310-322.

[21] SLATTERY A, WEN Z, TENBLAD P, et al. Automated self-optimization, intensification, and scale-up of photocatalysis in flow[J]. Science, 2024, 383(6681): eadj1817.

[22] HA T, LEE D, KWON Y, et al. AI-driven robotic chemist for autonomous synthesis of organic molecules[J]. Sci Adv, 2023, 9(44): eadj0461.

[23] BOIKO D A, MACKNIGHT R, KLINE B, et al. Autonomous chemical research with large language models[J]. Nature, 2023, 624(7992): 570-578.

[24] RUAN Y, LU C, XU N, et al. An automatic end-to-end chemical synthesis development platform powered by large language models[J]. Nat Commun, 2024, 15(1): 10160.

[25] DAI T, VIJAYAKRISHNAN S, SZCZYPIŃSKI F T, et al. Autonomous mobile robots for exploratory synthetic chemistry[J]. Nature, 2024, 635(8040): 890-897.

[26] GULATI G S, D’SILVA J P, LIU Y, et al. Profiling cell identity and tissue architecture with single-cell and spatial transcriptomics[J]. Nat Rev Mol Cell Biol, 2025, 26(1): 11-31.

[27] MOSHKOV N, BECKER T, YANG K, et al. Predicting compound activity from phenotypic profiles and chemical structures[J]. Nat Commun, 2023, 14(1): 1967.

[28] FREDIN HASLUM J, LARDEAU C H, KARLSSON J, et al. Cell Painting-based bioactivity prediction boosts high-throughput screening hit-rates and compound diversity[J]. Nat Commun, 2024, 15(1): 3470.

[29] BEGHIN A, GRENCI G, SAHNI G, et al. Automated high-speed 3D imaging of organoid cultures with multi-scale phenotypic quantification[J]. Nat Methods, 2022, 19(7): 881-892.

[30] MUKASHYAKA P, KUMAR P, MELLERT D J, et al. High-throughput deconvolution of 3D organoid dynamics at cellular resolution for cancer pharmacology with Cellos[J]. Nat Commun, 2023, 14(1): 8406.

[31] ONG H T, KARATAS E, POQUILLON T, et al. Digitalized organoids: Integrated pipeline for high-speed 3D analysis of organoid structures using multilevel segmentation and cellular topology[J]. Nat Methods, 2025, 22(6): 1343-1354.

[32] YU H, QIAN W, SONG Y, et al. PerturbNet predicts single-cell responses to unseen chemical and genetic perturbations[J]. Mol Syst Biol, 2025, 21(8): 960-982.

[33] GUO Y, ZHANG H, HU H, et al. Modelling drug-induced cellular perturbation responses with a biologically informed dual-branch transformer[J]. Nat Mach Intell, 2026, 8(1): 96-112.

[34] KAMYA P, OZEROV I V, PUN F W, et al. PandaOmics: An AI-driven platform for therapeutic target and biomarker discovery[J]. J Chem Inf Model, 2024, 64(10): 3961-3969.

[35] IVANENKOV Y A, POLYKOVSKIY D, BEZRUKOV D, et al. Chemistry42: An AI-driven platform for molecular design and optimization[J]. J Chem Inf Model, 2023, 63(3): 695-701.

[36] REN F, ALIPER A, CHEN J, et al. A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models[J]. Nat Biotechnol, 2025, 43(1): 63-75.

[37] XU Z, REN F, WANG P, et al. A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: A randomized phase 2a trial[J]. Nat Med, 2025, 31(8): 2602-2610.

[38] LI Y, ZHANG L, WANG Y, et al. Generative deep learning enables the discovery of a potent and selective RIPK1 inhibitor[J]. Nat Commun, 2022, 13(1): 6891.

[39] ZHANG L, LI Y, TIAN C, et al. From hit to lead: Structure-based optimization of novel selective inhibitors of receptor-interacting protein kinase 1 (RIPK1) for the treatment of inflammatory diseases[J]. J Med Chem, 2024, 67(1): 754-773.

[40] VLADIMER G, ALT I, SEHLKE R, et al. 23P Enriching for response: Patient selection criteria for A2AR inhibition by EXS-21546 through ex vivo modelling in primary patient material[J]. Immunooncol Technol, 2022, 16(s1): 100128.

[41] NARAYAN S, MCMAHON M, LIANOGLOU S, et al. Development of integrated stress response inhibitor RTX-117 as a treatment for neurodegenerative diseases (P9-11.028)[J]. Neurology, 2025, 104(7_Supplement_1): 4822.

[42] ÖZÇELIK R, GRISONI F. How evaluation choices distort the outcome of generative drug discovery[J]. J Cheminf, 2025, 17(1): 169.

[43] YIN Y, FU A, WANG L. Gaps between medical biology and AI drug discovery[J]. Drug Discov Today, 2025, 30(12): 104512.

[44] ABRAMSON J, ADLER J, DUNGER J, et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3[J]. Nature, 2024, 630(8016): 493-500.

基金资助

国家重点研发计划项目(2023YFF1204905)