I have run a MD simulation for a protein-ligand system using AMBER. There are multiple binding poses for the ligand during the simulation. I want to find out some representative binding poses of the ligand. How can I achieve this? I have tried the dbscan clustering by calculating the pair RMSD of every snapshots and I think it is not a good way.
To cluster ligand binding poses in a Molecular Dynamics (MD) simulation, various methods can be employed based on the search results:
1. Hierarchical Clustering : Utilize hierarchical clustering to group similar binding poses together based on structural similarity or other defined criteria [3][5].
2. Binding Energy-Based Clustering : Consider clustering based on predicted binding energy to identify distinct clusters of ligand poses [2].
3. Extended-Ensemble Docking : Employ extended-ensemble docking to probe dynamic variation in ligand binding and use clustering techniques to analyze the diverse conformations obtained [3].
4. Common Scaffold Clustering : Apply common scaffold clustering to identify ligand poses with a common binding orientation and cluster them accordingly for further analysis [5].
5. MCMC Process: Implement a Markov Chain Monte Carlo (MCMC) process to study ligand binding in multiple conformations and estimate the proportion of ligand binding in each position, aiding in clustering ligand poses effectively[4].
By utilizing these clustering techniques, researchers can effectively analyze and categorize ligand binding poses obtained from MD simulations, providing valuable insights into the interactions between ligands and proteins for drug design and other applications [1][2][3][4][5].
Citations:
[1] https://academic.oup.com/bioinformatics/article/34/5/770/4457357?login=false
[2] https://www.researchgate.net/figure/Binding-energy-based-clustering-of-docking-poses-of-ligands-A-1-and-B-2-in-the-ATP_fig5_325095935
[3] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8985516/
[4] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4434174/
[5] https://www.nature.com/articles/s41598-019-51455-8
What happens if you cluster the ligand’s atomic coordinates directly with DBSCAN or another method?
I would suggest running Surfaces (Article Surfaces: A software to quantify and visualize interactions ...
) for all poses - it is suitable for high throughput -, and from the results compare the matrices or maybe vectors of per-residue interactions. Using either euclidean or cosine distances, or even hierarchical clustering of these vectors, you should get a good comprehensive analysis of the poses.