Team capstone · 2026
Pocket-guided molecular docking
A team project testing whether predicted protein-binding pockets can guide docking toward more reliable ligand poses.
- Timeline
- May–July 2026
- Setting
- CMU Pre-College Computational Biology · team capstone
- Tools
- Python, PrankWeb, AutoDock Vina
The question
Can pocket predictions narrow the search?
Blind docking searches broadly across a protein. Our team asked whether predicting likely binding pockets first could improve the chance of recovering a successful pose.
The pipeline
Predict, dock, compare
We built a Python pipeline that used PrankWeb to predict pockets and AutoDock Vina to compare pocket-guided docking with blind docking. We benchmarked 97 protein–ligand complexes, then used 50 Astex complexes as a second evaluation set.
Team results
The figures below describe the team pipeline and benchmark, not individual performance.
Pocket-guided vs. blind docking
In our team’s 97-complex benchmark, pocket-guided docking recovered successful poses in 56% of cases, compared with 35% for blind docking, and rescued 27 cases.
Astex evaluation
In our team’s 50-complex Astex evaluation, the median RMSD was 1.36 Å, and 62% of complexes were under the 2 Å threshold.