An Accelerated Pocket Extraction and Evaluation Technique for Druggability

彬彬

Copyright is held by the author / owner(s).

SIGGRAPH Asia 2011, Hong Kong, China, December 12 – 15, 2011.

ISBN 978-1-4503-0807-6/11/0012

An Accelerated Pocket Extraction and Evaluation Technique for Druggability

Analysis with Protein Surfaces

Yukari Nakamura∗

Ochanomzu University

Ayaka Kaneko†

Ochanomizu Universiversity

Takayuki Itoh‡

Ochanomizu University

1 Introduction

Drugs act upon the concave portions of protein surfaces, so called

”pockets”. Discovery of well-shaped pockets on protein surfaces

is an active research topic. Several survey papers in this field report

many methods to search druggable pockets, which can be divided

in two major categories: geometric algorithms and/or energybased

methods. There are varieties of method-dependent geometric

descriptions of binding pockets such as depth, size, volume, and

amino acid composition, because there are no standard definitions

of what constitutes a pocket. We had a discussion with specialists

of drug discovery, and received the following suggestions: (1) It

should be reasonable to eliminate proteins which do not have wellshaped

pockets, before the chemical and energetic analysis phase.

(2) It is desirable to develop techniques that roughly but quickly

discover well-shaped pockets. Based on the discussion, this paper

presents a quick pocket extraction and evaluation technique. It applies

a mesh simplification technique to protein surfaces, extracts

concave portions from the simplified surfaces, and projects the portions

to the original surfaces. Finally, it evaluates the shapes of the

portions by comparing with preferably evaluated sample pockets.

2 Processing Flow

Protein Surface

Our technique uses protein surface datasets downloaded from the

database ”eF-site” (http://ef-site.hgc.jp/). We can freely obtain the

protein surfaces as triangular meshes in XML format, containing

vertices, edges, and triangles. Figure 1(Left) shows an example.

Mesh Simplification

Our technique aims to extract adequately-sized concave regions ignoring

smaller bumps. It applies a mesh simplification technique

using an implicit surface to get rough geometry by smoothing small

bumps. Our implementation generates a grid which surrounds the

protein surface, and then calculates the distance to the closest vertices

for each grid-point. Here, distances of the exterior grid-points

are positive, while distances of theinterior grid-points are negative.

It then generates an isosurface as the simplified protein surface, by

the Marching Cubes method with the zero-isovalue.

Concave Extraction

Let the position of a vertex P, and its normal vector N. Also, let

the position of the i-th vertex connected to the above vertex Pi. The

technique calculates N(P − Pi) with all of the connected vertices,

and assigns the attribute ”concave” to the vertex, if all the values

are negative. It then simply assigns the attribute ”concave” to the

triangles which are connected to one or more ”concave” vertices,

and treats the regions consisting of sets of adjacent ”concave” triangles

as pocket candidates. Figure 1(Right) shows an example of

pocket candidates on a simplified protein surface.

Concave Projection

Our technique projects the concave portions on a simplified mesh

onto the original mesh. Let triangles of the original mesh

∗e-mail:sincere@itolab.is.ocha.ac.jp

†e-mail:ayaka@itolab.is.ocha.ac.jp

‡e-mail:itot@is.ocha.ac.jp

Figure 1: (Left) Example of pocket surface. (Right) Example of

pocket extraction on a simplified mesh.

To = {t1, ..., tNO}, and triangles of simplified mesh Ts =

{s1, ..., sNS}. Our technique simply specifies si, which is the closest

to tj , and copies the attributes of si to tj .

Similarity Calculation with Sample Pockets

Our technique calculates the geometric feature values of the concave

portions by the following procedure. It evenly generates points

on the concave portions, calculates a histogram of their distribution,

and treats the histogram as the feature vector. The technique also

supposes that users collect sample concave portions which are truly

well-shaped and druggable. It calculates the feature values of the

sample concave portions and stores to a database as a preprocessing.

Our technique calculates the cosine similarity between the geometric

feature values of a new concave portion and stored sample

concave portion, and treats the maximum cosine value as the score

of the new concave portion.

3 Example and Future Work

Figure 2(Upper) shows an example with the protein ”1EZQ”, where

the surface contains 18,860 vertices and 37,316 triangles. Figure

2(Lower) shows another example with the protein ”1G1F”. Here,

red or orange portions are highly evaluated concaves. Small balls

around the surfaces are non-protein atoms remained during the

crystallization process of atom position measurement, where are

usually around truly druggable pockets. This result demonstrates

that our technique highly evaluated concave portions where nonprotein

atoms really remained. Computation time for pocket extraction

are quite small: the technique spent 0.203 seconds for mesh

simplification, 0.204 seconds for concave extraction on the rough

mesh, and 0.127 seconds for concave projection onto the original

mesh, respectively, to obtain the result with 1EZQ.

Figure 2: Examples of pocket extraction and evaluation result with

the protein ”1EZQ” and ”1G1F”.

tc

新年快乐，龙马精神，万事如意，百尺竿头，恭喜发财，年年有余。

C.R.CAN

无聊时可以刷屏幕 灌水 也可以试试 帖子的标题究竟可以写多长

奇

呵呵,真得不错哦!!

C.R.CAN

已阵亡的 蝶 随 风 舞 说过  偶尔按一下 CTRL A 会发现 世界还有另一面

C.R.CAN

水……生命之源……灌……

奇

很有心,部分已收录自用,谢谢

C.R.CAN

佩服，好多阿 ，哈哈

tc

很经典，很实用，学习了!

tc

楼主收集的可真全哦