1*Assistant Professor, Computer Science and Engineering, Parul University, Vadodara, Gujarat, India
Email: bhasha.anjaria21316@paruluniversity.ac.in
ORCID: 0000-0002-0562-4336
2Assistant Professor, Computer Science and Engineering, Parul University, Vadodara, Gujarat, India
Email: khushboo.trivedi21305@paruluniversity.ac.in
ORCID: 0009-0008-3197-4234
3Assistant Professor, Computer Science and Engineering, Parul University, Vadodara, Gujarat, India
Email: bhumi.shah19174@paruluniversity.ac.in
ORCID: 0009-0002-0868-2360
4Assistant Professor, Computer Science and Engineering, Parul University, Vadodara, Gujarat, India
Email: sujaya.bhattacharjee29571@paruluniversity.ac.in
ORCID: 0009-0008-9137-6447
5Assistant Professor, Computer Science and Engineering, Parul University, Vadodara, Gujarat, India
Email: limbachiya.arpi92@gmail.com
ORCID: 0009-0007-4911-7858
6Assistant Professor, Computer Science and Engineering, Parul University, Vadodara, Gujarat, India
Email: puvar.girirajsinh23266@paruluniversity.ac.in
ORCID: 0009-0004-0467-3146
Received: 19th Sep, 2025; Revised: 21st Oct, 2025; Accepted: 13th Nov, 2025; Available Online: 1st December, 2025
Artificial intelligence (AI)-based drug-target interaction (DTI) modeling has become a revolutionary approach to enhance drug discovery and to make drug delivery more efficient. A study has been proposed in this article to formulate an uncertainty aware transformer-based framework (SAFE-DTI Accuracy Boost) for predicting protein-ligand binding affinity using a BindingDB derived Kaggle dataset. The model combines dual transformer encoders, interactions between the models known as cross attention modeling, ranking regularization, and evidential regression to produce both affinity (detection) predictions and estimator of predictive uncertainty. Experimental evaluation on a test set of 60,000 interactions showed good regression performance (RMSE=0.9028, MAE=0.6359, Pearson r=0.8450). Binary classification at a pKd of 7.0 had an accuracy of 83%, which had balanced precision and recall between active and inactive classes. Analysis of the residuals supported the lack of any significant systematic bias and stable generalization. Compared to existing DTI models, the proposed framework is the first to take uncertainty quantification into account, which promotes reliable decision-making in the early screening stage and safety-awareness drug delivery optimization. The results indicate the promise of transformer-based evidential learning frameworks to minimize the burden on the experimental work, and support safer and more efficient therapeutic development strategies.
Keywords: Drug–Target Interaction; Transformer Architecture; Evidential Regression; BindingDB Dataset; Drug Delivery Optimization.
How to cite this article: Anjaria B, Trivedi KN, Shah BK, Bhattacharjee S, Limbachiya AM, Puvar G. AI-Powered Prediction Models For Enhancing Drug Delivery Efficacy And Safety. Int J Drug Deliv Technol. 2026;16(1): 708-718; DOI: 10.25258/ijddt.16.1.74
Source of support: Nil.
Conflict of interest: None