Citation: Meiru Liu, Yifei Xie, Shuang Li, Shiying Yang, Dezhi Yang, Zhengzheng Zhou, Li Zhang, Yang Lu, Guanhua Du. Multi-modal feature fusion graph neural networks for intelligent classification of pharmaceutical solid forms: Cocrystals, salts, and physical mixtures[J]. Chinese Chemical Letters, ;2026, 37(11): 112565. doi: 10.1016/j.cclet.2026.112565 shu

Multi-modal feature fusion graph neural networks for intelligent classification of pharmaceutical solid forms: Cocrystals, salts, and physical mixtures

Meiru Liu ^{a, 1} ,
Yifei Xie ^{a, 1} ,
Shuang Li ^a ,
Shiying Yang ^a ,
Dezhi Yang ^{a, *,} ,
Zhengzheng Zhou ^{b, *,} ,
Li Zhang ^{a, *,} ,
Yang Lu ^a ,
Guanhua Du ^a

a.
Beijing Key Laboratory of Innovative Drug Discovery and Polymorphic Research for Cerebrovascular Diseases, Institute of Materia Medica, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100050, China
b.
NMPA Key Laboratory for Safety Evaluation of Cosmetics, Guangdong Provincial Key Laboratory of Tropical Disease Research, Department of Hygiene Inspection & Quarantine Science, School of Public Health, Southern Medical University, Guangzhou 510515, China

ydz@imm.ac.cn

zhouzz418@smu.edu.cn

zhangl@imm.ac.cn

Received Date: 31 July 2025
Revised Date: 16 January 2026
Accepted Date: 1 March 2026
Available Online: 2 March 2026

Figures(5)

Pharmaceutical solid form engineering through co-crystallization represents a transformative strategy for enhancing drug bioavailability and physicochemical properties, yet remains hindered by labor-intensive experimental screening and the notable absence of computational models capable of simultaneously classifying co-crystals, salts, and physical mixtures. Here, we present an intelligent multi-modal graph neural network (GNN) framework that synergistically integrates molecular graph representations, molecular fingerprints, and quantum chemical descriptors to enable automated three-class classification of pharmaceutical solid forms, addressing a critical gap in current computational pharmaceutical research. Our innovative feature fusion strategy combines topological molecular graphs with traditional molecular fingerprints and computationally-derived quantum chemical properties, creating a comprehensive multi-dimensional representation of pharmaceutical compounds. The developed integrated model achieves exceptional classification accuracy exceeding 98% across all evaluation phases, demonstrating the superior performance of combined structural, topological, and electronic feature representations in complex multi-component pharmaceutical systems. Rigorous validation using three independent active pharmaceutical ingredient datasets, theophylline, ligustrazine, and piperazine, yields outstanding success rates, confirming the model's remarkable generalizability and practical applicability across diverse chemical spaces. Most significantly, our artificial intelligence (AI)-driven multi-modal predictions successfully guided the experimental synthesis of two novel co-crystalline phases: Emodin-ligustrazine and emodin-metformin, validating the predictive power of integrated molecular feature representations in real-world pharmaceutical development.
- Pharmaceutical solid form,
- Graph neural networks,
- Multi-class classification,
- Quantum chemistry,
- Co-crystallization,
- Emodin
1. [1]
  D. Yan, D.K. Bučr, A. Delori, et al., Chem. Eur. J. 19 (2013) 8213–8219. doi: 10.1002/chem.201203810
2. [2]
  D. Yan, Chem. Eur. J. 21 (2015) 4880–4896. doi: 10.1002/chem.201405456
3. [3]
  S. Li, B. Lu, X. Fang, et al., Angew. Chem. Int. Ed. 59 (2020) 22623–22630. doi: 10.1002/anie.202009714
4. [4]
  M. Dai, Z. Qi, D. Yan, Angew. Chem. Int. Ed. 64 (2025) e202420139. doi: 10.1002/anie.202420139
5. [5]
  M. Guo, X. Sun, J. Chen, et al., Acta Pharm. Sin. B 11 (8) (2021) 2537–2564. doi: 10.1016/j.apsb.2021.03.030
6. [6]
  Y. Lu, Y. Tang, H. Lin, et al., Chin. Chem. Lett. 29 (2018) 1541–1543. doi: 10.1016/j.cclet.2017.12.021
7. [7]
  C. Xing, B. Zhou, D. Yan, CCS Chem. 5 (2023) 2866–2876. doi: 10.31635/ccschem.023.202202605
8. [8]
  H. Liu, H.C.S. Chan, L. Zhang, et al., Chin. Chem. Lett. 34 (2023) 108057. doi: 10.1016/j.cclet.2022.108057
9. [9]
  J. Li, H. Liu, L. Zhang, et al., Chin. Chem. Lett. 33 (2022) 4069–4073. doi: 10.1016/j.cclet.2022.03.090
10. [10]
  R. Nathi, N.V.D.P. Ketha, L.P. Kowtarapu, et al., Pharm. Sci. Adv. 1 (2023) 100018. doi: 10.1016/j.pscia.2023.100018
11. [11]
  P. Roy, A. Ghosh, CrystEngComm 22 (2020) 6958–6974. doi: 10.1039/d0ce01276a
12. [12]
  K.O. Oyediran, I.M. Cardoso-Daodu, P.O. Bassey, et al., Pharm. Sci. Adv. 3 (2025) 100079. doi: 10.1016/j.pscia.2025.100079
13. [13]
  E. Pindelska, A. Sokal, W. Kolodziejski, Adv. Drug Deliv. Rev. 117 (2017) 11–146.
14. [14]
  L. Bodart, J. Body, N. Tumanov, et al., Cryst. Growth Des. 23 (2023) 7811–7820. doi: 10.1021/acs.cgd.3c00626
15. [15]
  J.R. Štoček, O. Socha, I. Císařová, et al., J. Am. Chem. Soc. 144 (2022) 7111–7116. doi: 10.1021/jacs.1c10885
16. [16]
  Y. Deng, S. Liu, Y. Jiang, et al., Pharmaceutics 15 (2023) 2174. doi: 10.3390/pharmaceutics15092174
17. [17]
  P.M. Bhatt, N.V. Ravindra, R. Banerjee, et al., Chem. Commun. 8 (2005) 1073–1075. doi: 10.1039/b416137h
18. [18]
  A.A. Ahangar, H. Qadri, A.A. Malik, et al., Mol. Pharm. 20 (2023) 3471–3483. doi: 10.1021/acs.molpharmaceut.3c00087
19. [19]
  Y. Jiang, Z. Yang, J. Guo, H. Li, et al., Nat. Commun. 12 (2021) 5950. doi: 10.1038/s41467-021-26226-7
20. [20]
  J.J. Devogelaer, H. Meekes, P. Tinnemans, et al., Angew. Chem. Int. Ed. 59 (2020) 21711–21718. doi: 10.1002/anie.202009467
21. [21]
  M.A. Ghanavati, S. Rohani, Cryst. Growth Des. 25 (2025) 2717–2729. doi: 10.1021/acs.cgd.5c00347
22. [22]
  S. Liu, W. Nie, C. Wang, et al., Nat. Mach. Intell. 5 (2023) 1447–1457. doi: 10.1038/s42256-023-00759-6
23. [23]
  T. Wu, Y. Tang, Q. Sun, et al., Comput. Biol. Bioinform. 20 (2023) 3044–3055. doi: 10.1109/tcbb.2023.3253862
24. [24]
  R. Chen, C. Li, L. Wang, et al., Inf. Fusion 115 (2025) 102784. doi: 10.1016/j.inffus.2024.102784
25. [25]
  J. Yang, Y. Cai, K. Zhao, et al., Drug Discov. Today 27 (2022) 103356. doi: 10.1016/j.drudis.2022.103356
26. [26]
  A. Cereto-Massagué, M.J. Ojeda, C. Valls, et al., Methods 71 (2015) 58–63. doi: 10.1016/j.ymeth.2014.08.005
27. [27]
  L. Zheng, J. Cao, L. Jing, et al., Pharm. Sci. Adv. 4 (2026) 100100. doi: 10.1016/j.pscia.2025.100100
28. [28]
  D. Yang, H. Wang, Q. Liu, et al., Chin. Chem. Lett. 33 (2022) 3207–3211. doi: 10.1016/j.cclet.2021.10.012
29. [29]
  H. Wang, D. Yang, W. Zhang, et al., Chin. Chem. Lett. 34 (2023) 107258. doi: 10.1016/j.cclet.2022.02.063
30. [30]
  H.Y.I. Lam, R. Pincket, H. Han, et al., Nat. Mach. Intell. 5 (2023) 754–764. doi: 10.1038/s42256-023-00683-9
31. [31]
  K. Atz, F. Grisoni, G. Schneider, et al., Nat. Mach. Intell. 3 (2021) 1023–1032. doi: 10.1038/s42256-021-00418-8
32. [32]
  Z.A. Rollins, A.C. Cheng, E. Metwally, J. Cheminform. 16 (2024) 56. doi: 10.1186/s13321-024-00846-9
33. [33]
  W. Ahmad, H. Tayara, K.T. Chong, ACS Omega 8 (2023) 3236–3244. doi: 10.1021/acsomega.2c06702
34. [34]
  Q. Yang, Y. Li, J.D. Yang, et al., Angew. Chem. Int. Ed. 59 (2020) 19282–19291. doi: 10.1002/anie.202008528
35. [35]
  M. Abadi, P. Barham, J. Chen, et al., OSDI 16 (2016) 265–283.
36. [36]
  A. Kensert, G. Desmet, D. Cabooter, J. Comput. Aided Mol. Des. 39 (2025) 3. doi: 10.1007/s10822-024-00578-w
37. [37]
  F. Pedregosa, G. Varoquaux, A. Gramfort, et al., J. Mach. Learn. Res. 12 (2011) 2825–2830.
38. [38]
  S. Fu, W. Liu, K. Zhang, Y. Zhou, et al., Neurocomputing 461 (2021) 63–76. doi: 10.1016/j.neucom.2021.07.048
39. [39]
  V. Fung, J. Zhang, E. Juarez, et al., npj Comput. Mater. 7 (2021) 84. doi: 10.1038/s41524-021-00554-0
40. [40]
  S. Ryu, Y. Kwon, W.Y. Kim, Chem. Sci. 10 (2019) 8438–8446. doi: 10.1039/c9sc01992h
41. [41]
  X. Ye, Q. Guan, W. Luo, et al., Pattern Recognit. 128 (2022) 108659. doi: 10.1016/j.patcog.2022.108659
42. [42]
  B. Lu, S. Liu, D. Yan, Chin. Chem. Lett. 30 (2019) 1908–1922. doi: 10.1016/j.cclet.2019.09.012
43. [43]
  S. Li, Y. Lin, D. Yan, J. Mater. Chem. C 4 (2016) 2527–2534. doi: 10.1039/C6TC00067C
44. [44]
  C.R. Groom, I.J. Bruno, M.P. Lightfoot, et al., Acta Crystallogr. Sect. B Struct. Sci. Cryst. Eng. Mat. 72 (2016) 171–179. doi: 10.1107/S2052520616003954
45. [45]
  D. Yang, L. Wang, P. Yuan, et al., Chin. Chem. Lett. 34 (2023) 107964. doi: 10.1016/j.cclet.2022.107964
46. [46]
  G. Landrum, GitHub repository, accessed 30 June 2025, https://github.com/rdkit/rdkit.
47. [47]
  A.K. Rappe, C.J. Casewit, K.S. Colwell, et al., J. Am. Chem. Soc. 114 (1992) 10024–10035. doi: 10.1021/ja00051a040
48. [48]
  N.M. O'Boyle, M. Banck, C.A. James, et al., J. Cheminform. 3 (2011) 33. doi: 10.1186/1758-2946-3-33
49. [49]
  T. Zeng, J. Li, R. Wu, Pharm. Sci. Adv. 2 (2024) 100050. doi: 10.1016/j.pscia.2024.100050
50. [50]
  A. Capecchi, D. Probst, J.L. Reymond, J. Cheminform. 12 (2020) 43. doi: 10.1186/s13321-020-00445-4
51. [51]
  P. Gedeck, B. Rohde, C. Bartels, J. Chem Inf. Model. 46 (2006) 1924–1936. doi: 10.1021/ci050413p
52. [52]
  J.L. Durant, B.A. Leland, D.R. Henry, et al., J. Chem. Inf. Model 42 (2002) 1273–1280.
53. [53]
  H.L. Morgan, J. Chem. Doc. 5 (1965) 107–113. doi: 10.1021/c160017a018
54. [54]
  M. Frisch, G. Trucks, H. Schlegel, et al., Gaussian, 16, Gaussian, Inc., Wallingford, CT, 2016 Revision A. 03.
55. [55]
  T. Lu, F. Chen, J. Comput. Chem. 33 (2012) 580–592. doi: 10.1002/jcc.22885
56. [56]
  T. Lu, J. Chem. Phys. 161 (2024) 082503. doi: 10.1063/5.0216272
57. [57]
  L. Wang, L. Zhao, X. Liu, et al., Environ. Sci. Technol. 55 (2021) 149958–149967.
58. [58]
  V. Saini, A. Sharma, D. Nivatia, Phys. Chem. Chem. Phys. 24 (2022) 1821–1829. doi: 10.1039/d1cp05072a
1. [1]
  Shuang Li , Penghui Yuan , Xinyi Zhang , Meiru Liu , Dezhi Yang , Linglei Kong , Li Zhang , Yang Lu , Guanhua Du . Revolutionizing sepsis therapy: Machine learning-driven co-crystallization reveals emodin's therapeutic potential. Chinese Chemical Letters, 2026, 37(2): 111289-. doi: 10.1016/j.cclet.2025.111289
2. [2]
  Youcheng Hu , Guichan Li , Chunli Qi , Xiaoyan Lin , Xingwang Zhang , Hao Wang . Macrophage membrane-engineered nanovesicles for targeted delivery of emodin in ulcerative colitis via chemotactic homing. Chinese Chemical Letters, 2026, 37(7): 111727-. doi: 10.1016/j.cclet.2025.111727
3. [3]
  Jun Lu , Jinrui Yan , Yaohao Guo , Junjie Qiu , Shuangliang Zhao , Bo Bao . Controlling solid form and crystal habit of triphenylmethanol by antisolvent crystallization in a microfluidic device. Chinese Chemical Letters, 2024, 35(4): 108876-. doi: 10.1016/j.cclet.2023.108876
4. [4]
  Jia Zhou . Constructing Potential Energy Surface of Water Molecule by Quantum Chemistry and Machine Learning: Introduction to a Comprehensive Computational Chemistry Experiment. University Chemistry, 2024, 39(3): 351-358. doi: 10.3866/PKU.DXHX202309060
5. [5]
  Cuijuan Xuan , Jie Wang , Li Yin , Xiuling Xu , Fengyu Du , Lubin Xu , Liangyu Gong . Construction and practice of a “dual-perspective three-dimensional” experimental teaching model enabled by AI-quantum chemistry synergy in the context of new agricultural science. University Chemistry, 2026, 41(7): 67-75. doi: 10.12461/PKU.DXHX202506031
6. [6]
  Guoxian Zhu , Jing Chen , Rongkai Pan . Enhancing the Teaching Quality of Atomic Structure: Insights and Strategies. University Chemistry, 2024, 39(3): 376-383. doi: 10.3866/PKU.DXHX202305027
7. [7]
  Dongju Zhang , Rongxiu Zhu . Construction of Ideological and Political Education in Quantum Chemistry Course: Several Teaching Cases to Reveal the Universal Connection of Things. University Chemistry, 2024, 39(7): 272-277. doi: 10.3866/PKU.DXHX202311032
8. [8]
  Xiumei LI , Yanju HUANG , Bo LIU , Yaru PAN . Syntheses, crystal structures, and quantum chemistry calculation of two Ni(Ⅱ) coordination polymers. Chinese Journal of Inorganic Chemistry, 2024, 40(10): 2031-2039. doi: 10.11862/CJIC.20240109
9. [9]
  Hong-Li Long , Hong-Jie Peng . Lithium-bond chemistry enlightens 600 Wh/kg solid-state batteries. Chinese Chemical Letters, 2026, 37(2): 112045-. doi: 10.1016/j.cclet.2025.112045
10. [10]
  Jia-Xin Wu , Zheng Yin , Ming-Hua Zeng . Multi-phase evolution of MOFs involving crystal, liquid and glass: new dynamic chemistry. Chinese Journal of Structural Chemistry, 2025, 44(10): 100710-100710. doi: 10.1016/j.cjsc.2025.100710
11. [11]
  Shu-Ran Xu , Fang-Xing Xiao . Metal halide perovskites quantum dots: Synthesis, and modification strategies for solar CO2 conversion. Chinese Journal of Structural Chemistry, 2023, 42(12): 100173-100173. doi: 10.1016/j.cjsc.2023.100173
12. [12]
  Yinghua Lv , Yuchao Lu , Chenxi Hou , Jun Li , Zelong Chen , Aihua Zhen , Jiaming Ren , Yuxin Pei , Tony D James , Lan Chen , Yanjie Jia , Danhong Shang , Pengkun He , Zhongfu Wang , Zhichao Pei . Supramolecular glycosyl-nanoproteins based on host-guest chemistry for targeted co-delivery in vivo. Chinese Chemical Letters, 2026, 37(11): 112185-. doi: 10.1016/j.cclet.2025.112185
13. [13]
  Haowen Li , Hongying Hou , Dai-Huo Liu , Bao Li , Dongmei Dai , Bao Wang , Mengmin Jia , Zhuangzhuang Zhang , Liang Wang , Yaru Qiao , Canhui Wu , Huihui Zhu , Pengyao Yan . A designed flexible solid-state electrolyte with rich hydrogen-bonded networks from TPU-PEGDA/LLZTO for Li metal batteries. Chinese Chemical Letters, 2026, 37(2): 111020-. doi: 10.1016/j.cclet.2025.111020
14. [14]
  Yong Zhang , Jiangnan Yu , Chao Liang , Zhaomin Li , Xueqin Li . Mimicking tollgate-highway systems by cellulose-MXene networks in MMMs for enhancing CO₂ separation. Chinese Chemical Letters, 2026, 37(6): 112065-. doi: 10.1016/j.cclet.2025.112065
15. [15]
  Zhijie Zhang , Xun Li , Huiling Tang , Junhao Wu , Chunxia Yao , Kui Li . Cs₂CuBr₄ perovskite quantum dots confined in mesoporous CuO framework as a p-n type S-scheme heterojunction for efficient CO₂ photoconversion. Chinese Chemical Letters, 2024, 35(11): 109700-. doi: 10.1016/j.cclet.2024.109700
16. [16]
  Ziyu Yao , Huan Yang , Luyang Zuo , Fang Wang . Both promoting tetracycline decomposition in combination with CO₂ reduction for flower-like MoS₂ modified by Ti₃C₂ MXene quantum dots. Chinese Chemical Letters, 2026, 37(9): 111595-. doi: 10.1016/j.cclet.2025.111595
17. [17]
  Shehla Khalid , Muhammad Bilal , Nasir Rasool , Muhammad Imran . Photochemical reactions as synthetic tool for pharmaceutical industries. Chinese Chemical Letters, 2024, 35(9): 109498-. doi: 10.1016/j.cclet.2024.109498
18. [18]
  Wei Chen , Pieter Cnudde . A minireview to ketene chemistry in zeolite catalysis. Chinese Journal of Structural Chemistry, 2024, 43(11): 100412-100412. doi: 10.1016/j.cjsc.2024.100412
19. [19]
  Siting Cai , Xiang Chen , Shuli Wang , Xinqin Liao , Zhong Chen , Yue Lin . Silica coating of quantum dots and their applications in optoelectronic fields. Chinese Chemical Letters, 2025, 36(6): 110798-. doi: 10.1016/j.cclet.2024.110798
20. [20]
  Peiwen Liu , Fang Zhao , Jing Zhang , Yunpeng Bai , Jinxing Ye , Bo Bao , Xinggui Zhou , Li Zhang , Changlu Zhou , Xinhai Yu , Peng Zuo , Jianye Xia , Lian Cen , Yangyang Yang , Guoyue Shi , Lin Xu , Weiping Zhu , Yufang Xu , Xuhong Qian . Micro/nano flow chemistry by Beyond Limits Manufacturing. Chinese Chemical Letters, 2024, 35(5): 109020-. doi: 10.1016/j.cclet.2023.109020

Get Citation

PDF

XML

Metrics

PDF Downloads(0)
Abstract views(39)
HTML views(5)

通讯作者: 陈斌, bchen63@163.com

1.
沈阳化工大学材料科学与工程学院沈阳 110142