COS

URC Undergraduate Research

A Clustering Algorithm for Molecular Structures:

Application on the Met-Enkephalin Peptide Ruxi Xiang1, M. Jennifer Van1, Mahmoud Namazi2,3, Estela Blaisten-Barojas 4,5,*, and Amarda Shehu1,6,*

1Dept. of Computer Science, 2Dept. of Mathematical Sciences, 3Dept. of Electrical and Computer Engineering, 4Computational Materials

Science Center, and 5School of Physics, Astronomy, and Computational Sciences, 6Dept. of Bioengineering

George Mason University, Fairfax, VA 22030 *[blaisten or amarda]@gmu.edu

Abstract

Met-enk is an endogeneous peptide that mediates pain

and dependence on opioids [1].

It flexes its structure to bind different opioid receptors

Wet-laboratory techniques have revealed a few structural

states of met-enk [2].

Research Objective: provide a comprehensive view of the

structure space of met-enk through a variety of

computational methods.

Project team of two faculty and three undergraduate

student researchers.

My role: design of an algorithm to cluster computed

structures of met-enk and so automate detection of

thermodynamically-stable structural states.

Methodology Continued

Flowchart of algorithm Distance function

D 𝑉1, 𝑉2 =1

𝑛 |𝑉1𝑖 − 𝑉2𝑖|

𝑛

𝑖=1

Φ1

Ψ1

ω1

…

Φi

Ψi

Ωi

..

Φn

Ψn

V1 V2

Φ1

Ψ1

ω1

…

Φi

Ψi

Ωi

..

Φn

Ψn

Methodology

Clustering algorithm adapts the known SPICKER

algorithm used to cluster computed decoys in de novo

protein structure prediction. [3]

Original implementation uses a computationally-

expensive dissimilarity measures obtained after optimal

superimposition of two structures.

Our adaptation is to integrate a fast, novel angular-based

distance function

Results Clustering of Wet-lab Structures

Clustering of Structures obtained via Molecular

Dynamics with AMBER force fields [4]

Clustering of Structures obtained via Basin Hopping and

Rosetta energy function [5]

Cluster 1 Cluster 2

15 5

2LWC 1PLX

Cluster 1 Cluster 2

57 23

Cl. 1 Cl. 2 Cl. 3 Cl.4

57 21 1 1

1PLW

Cluster 4 Cluster 5

649 641

Cluster 4 Cluster 5

647 643

References [1] Koneru, A, Satyanarayana S, and Rizwan S. Global J of Pharmacology

3 (2009): 149-153.

[2] Graham et al. Biopolymers 32.12 (1992): 1755-1764.

[3] Zhang Y and Skolnick J. J Comp Chem 25. (2004): 865-871.

[4] Dai Y. and Blaisten-Barojas E. J. Chem. Phys. 133, (2010), 034905

[5] Olson B. and Shehu A. Proteome Sci 11 (2013), S1.

Acknowledgements

The Thomas F. and Kate Miller Jeffress Memorial

Trust Award to AS and EB

Cluster 1 Cluster 2 Cluster 3

1023 920 802

Cluster 1 Cluster 2 Cluster 3

4006 1629 708

Conclusion

Clusters highlight met-enk populates diverse states

Algorithm can cluster >10,000 structures in < 1 minute of CPU time.

Algorithm is generally applicable for analysis of molecular structure data.

Correspondence to be established to determine which wet-lab states are

reproduced and which ones are novel and discovered in silico