Proteolysis-Targeting Chimera (PROTAC) Design using Integrated Artificial Intelligence (AI) and Quantitative Systems Pharmacology (QSP) Model.
Jina Kim (1), Sungwoo Goo (1), Sangkeun Jung (1,2)*, Jung-woo Chae (1,3)*, Jae-mun Choi (4)*, Hwi-yeol Yun (1,3)*
(1) Department of Bio-AI convergence, Chungnam National University, DaeJeon, Republic of Korea (2) Department of Computer Science and Engineering, Chungnam National University, DaeJeon, Republic of Korea (3) College of Pharmacy, Chungnam National University, DaeJeon, Republic of Korea (4) Calici Co., LTD, Daejeon, Republic of Korea *Those of authors contributed equally as correspondence.
Objectives:
Proteolysis-Targeting Chimera (PROTAC) is new type of bifunctional engineered molecule what it allows to bring the target protein and the E3 ubiquitin ligase, leading to ubiquitination and subsequent degradation of the protein of interest (POI) by the proteasome. To optimal engineer the PROTAC, it is the major challenge to screen and select proper PROTAC combination in evaluating the degree of degradation. Thus, the main purpose of this study is to design a new strategy that it can predict the optimal PROTAC combination using the integration of the Structure-based artificial intelligence (AI) model and quantitative systems pharmacology (QSP) model. Ultimately, our approach has the potential to significantly streamline the PROTAC development process, leading to the discovery of more effective therapeutic agent.
Methods:
The overall workflow was conducted by applying the Kd (the complex dissociation constants) values between the POI, PROTAC, and E3 ubiquitin ligases predicted by the AI model to the QSP model referring to the study by Haid et al. [1], which describes the kinetics between them. For the PROTAC combinatorial screening, we found 83 combinations of E3 ubiquitination ligases with E3 ligands and 1038 combinations between POIs and warheads in the PROTAC DB [2], the PROTAC experimental database, and converted them into three-dimensional structures using AlphaFold2 and OpenBabel. In addition to the Kd value obtained through the AI model, E3 ubiquitin ligase abundance data obtained from the protein abundance database (PAXdb) and half-life data of POIs obtained from Proteomics DB were used as parameters of the QSP model. These parameters were applied to the QSP model to screen the degradation efficiency of PROTAC combinations with degradation parameters such as DC50 (the PROTAC concentration that produces half-maximal degradation) and Dmax (the maximal extent of degradation).
Results:
Our PROTAC Design Model is designed to evaluate the binding stability of ternary complexes based on their kd values, as well as calculate the DC50 and Dmax values of POIs by inputting the Simplified Molecular Input Line Entry System (SMILES) for chemicals and amino acid sequence information for proteins. In order to evaluate the reliability of the AI model, the predicted accuracy was evaluated by comparing the measured kd value and the predicted Kd value using the AI model. As a result, it performed with a Root Mean Squared Error (RMSE) of 1.748, Mean Absolute Error (MAE) of 1.403, and R-squared (R2) of 0.194. To evaluate the prediction accuracy of the QSP model, degradation data of 18 PROTAC combinations and parameter data required for the QSP model were used as input values for the QSP model [1,3,4,5]. Prediction accuracy was obtained by comparing the difference between the predicted DC50 and Dmax using the reference data and the measured DC50 and Dmax. As a result, DC50 showed RMSE 1.7957, MAE 1.4791, Pearson correlation 0.8049, and Dmax showed RMSE 9.6606, MAE 7.6331, Pearson correlation 0.9512. In order to select potential PROTAC combinations and prioritize among them what we screened, we compared the degradation levels of ARV-110 and ARV-471, the candidate of PROTAC drugs in latest stage. The average DC50 and Dmax of both drugs were 1.4 (nM) and 96 (%) [6,7], respectively, and those values was used as criteria for screening out of a PROTAC drug that has the potential ability as PROTAC candidates. As we evaluated PROTAC combinations containing 8 E3 ubiquitin ligases including Cereblon and von Hippel-Lindau and 91 POIs including androgen receptor and estrogen receptor, 35919 potential PROTAC combinations were not only screened out as potential candidates in total 40410, but the priority among those combinations also be described based on predicted DC50 and Dmax.
Conclusions:
Our PROTACs design is a leading attempt to integrate AI and QSP into the field of PROTACs. It should be helpful to make candidates for optimal PROTAC combination based on highly efficient screening results as well as suggestion the PROTAC research paradigm for combining AI and QSP. Along with the refinement of the PROTAC design, further approaches will need to be taken to integrate experimental results to create sophisticated models.
References:
- Haid, R. T. U., & Reichel, A. (2023). A Mechanistic Pharmacodynamic Modeling Framework for the Assessment and Optimization of Proteolysis Targeting Chimeras (PROTACs). Pharmaceutics, 15(1), 195.
- Weng, G., Shen, C., Cao, D., Gao, J., Dong, X., He, Q., ... & Hou, T. (2021). PROTAC-DB: an online database of PROTACs. Nucleic acids research, 49(D1), D1381-D1387.
- Zorba, A., Nguyen, C., Xu, Y., Starr, J., Borzilleri, K., Smith, J., ... & Calabrese, M. F. (2018). Delineating the role of cooperativity in the design of potent PROTACs for BTK. Proceedings of the National Academy of Sciences, 115(31), E7285-E7292.
- Guo, W. H., Qi, X., Yu, X., Liu, Y., Chung, C. I., Bai, F., ... & Wang, J. (2020). Enhancing intracellular accumulation and target engagement of PROTACs with reversible covalent chemistry. Nature communications, 11(1), 4268.
- Bradshaw, J. M., McFarland, J. M., Paavilainen, V. O., Bisconte, A., Tam, D., Phan, V. T., ... & Taunton, J. (2015). Prolonged and tunable residence time using reversible covalent kinase inhibitors. Nature chemical biology, 11(7), 525-531.
- Snyder, L. B., Neklesa, T. K., Chen, X., Dong, H., Ferraro, C., Gordon, D. A., ... & Taylor, I. (2021). Discovery of ARV-110, a first in class androgen receptor degrading PROTAC for the treatment of men with metastatic castration resistant prostate cancer. Cancer Research, 81(13_Supplement), 43-43.
- Flanagan, J. J., Qian, Y., Gough, S. M., Andreoli, M., Bookbinder, M., Cadelina, G., ... & Houston, J. ARV-471, an oral estrogen receptor PROTAC degrader for breast cancer, Proceedings of the 2018 San Antonio Breast Cancer Symposium, San Antonio, TX, Philadelphia (PA): AACR, 2018. Cancer. Res., 79.
