Automated Drug Design Pipeline Integrating Deep Learning-Based Docking and Molecular Dynamics

Author

• Tony Enrique Astuhuaman Davila, Missouri State UniversityFollow

Date of Graduation

Spring 2026

Degree

Master of Natural and Applied Science in Physics

Department

Physics, Astronomy and Materials Science

Committee Chair

Ridwan Sakidja

Abstract

Computational drug screening pipelines that rely solely on molecular docking for candidate prioritization are vulnerable to false positives because static scoring cannot distinguish geometrically plausible poses from dynamically stable ones. This limitation is especially critical where resistance mutations emerge rapidly, requiring reliable computational triage against new or mutated targets. This study presents a staged, reproducible screening protocol. It combines diffusion-based ligand generation, medicinal-chemistry filtering, learned docking, pocket-level refinement, consensus ranking, and explicit-solvent MD. As a case study, the protocol is demonstrated against PDB 4YY8, a Plasmodium falciparum kelch-domain crystal structure associated with artemisinin-resistant malaria. A staged evidence design with progressively increasing replicate depth and simulation horizon reduced 300 generated molecules to three finalists. Two candidates were classified as retained under uncertainty-aware ranking, while one was deprioritized as a docking-stage false positive despite competitive static scores. Cross-stage analysis confirmed that ranking conclusions changed materially with simulation depth. The results demonstrate that staged AI and MD refinement provides more reliable screening-level triage than docking alone. All conclusions are limited to computational screening; no therapeutic efficacy claims are made.

Keywords

computational drug design, molecular dynamics, diffusion models, virtual screening, structure-based drug design, uncertainty-aware ranking, malaria, drug resistance, Plasmodium falciparum, reproducible workflow

Subject Categories

Artificial Intelligence and Robotics | Bioinformatics | Biophysics

Copyright

© Tony Enrique Astuhuaman Davila

Recommended Citation

Astuhuaman Davila, Tony Enrique, "Automated Drug Design Pipeline Integrating Deep Learning-Based Docking and Molecular Dynamics" (2026). Graduate Theses/Dissertations. 4176.
https://bearworks.missouristate.edu/theses/4176