Embed Size (px)
Citation preview
1
Spresso:An ultrafast compound pre-screening method based on compound decomposition
Keisuke Yanagisawa1,2 Shunta Komine2,3
Shogo D. Suzuki2,3 Masahito Ohue1,3,4
Takashi Ishida1,2,3,4 Yutaka Akiyama1,2,3,4
1. Dept. CS, School of Computing, Tokyo Tech.2. Education Academy of Computational Life Sciences (ACLS), Tokyo Tech.3. Dept. CS, Graduate School of Information Science and Engineering, Tokyo Tech.4. Advanced Computational Drug Discovery Unit (ACDD), Tokyo Tech.
© 2016 Keisuke YANAGISAWA* The software was renamed from ESPRESSO to Spresso (same pronunciation) in October 2016.
2 © 2016 Keisuke YANAGISAWA
Docking
> 107 entries ~ 105 entries ~ 103 entries
Speedy PRE-Screening methodwith Segmented cOmpounds
( http://www.bi.cs.titech.ac.jp/spresso/ )
Background
Docking-based virtual screening
3 © 2016 Keisuke YANAGISAWA
Virtual screening
Compound DB Drug candidates
Docking calculation
A protein & a compound A conformation
Background
Conformation search
4 © 2016 Keisuke YANAGISAWA
Translation3 dimensions
Rotation3 dimensions
Internal rotationN dimensions
O
O
OH
OO
OH
Problem: Computationally expensive
Background
Compound pre-screening
5
Decreasing calculation with pre-screening
Compound DB Drug candidates
Ex: 10,000,000 100,000 1,000x 1/100
compoundscompoundscompounds
© 2016 Keisuke YANAGISAWA
DockingPre-screening
Background
Existing pre-screening methods
6
Pre-screening
Ligand-based Structure-based
ML1), Shape matching2) etc. Rough docking (Glide HTVS)3,4)
No bias toward known actives
Insufficient speed(1 sec / compound =
1 CPU year for ZINC DB)
1) H. Geppert, et al. J Chem Inf Model, 2010. 2) A. Vuorinen, et al. J Med Chem, 2014.
3) S. B. Mirza, et al. J Mol Graph Model, 2016.4) A. Grover, et al. Biochim Biophys Acta, 2012.
© 2016 Keisuke YANAGISAWA
Requiring known compoundsresulting in difficulty finding novel drug candidates
Light weight calculation
ON
NH+
N
O
ClO
-O
O
N
N
N-
N
NN
O-O
O
O
+
HN
N
NH2
N
N
O
O
O
Background
Our approach
7
Ligand-based Structure-based Spressowo/ known compound
pre-screening speed
© 2016 Keisuke YANAGISAWA
• Structure-based
• Ultrafast compared to existing method
Ideas for acceleration
8
Idea I. Compound decomposition
Idea II. Rough compound evaluation
Method
O
O
OH
O
O
OH
Fragment
score ＝ 𝑤𝑤score = 𝑥𝑥 score = y score = 𝑧𝑧
Compound
© 2016 Keisuke YANAGISAWA
Idea I. Compound decomposition
9
Method
O
O
OH
Creating fragments without any rotatable bond1)
OO
OH
H
H
1) S. Komine et al., IPSJ SIG Technical Report, 2015-BIO-42, 2015.
© 2016 Keisuke YANAGISAWA
Compound Fragments
0200,000400,000600,000800,000
1,000,000
Another benefit of decomposition
10
Method
Sharing of fragment docking results for duplication
Decomposition result compounds: ZINC “drugs now” subset (10,639,555 entries)
O
Cl
OF
HOO
NH
Example of duplicated fragments
#St
ruct
ure
varia
tions
#Compounds
© 2016 Keisuke YANAGISAWA
compound
fragment
150
variation
Idea II. Rough compound evaluation
11
Method
Compound evaluation without re-construction
Fragments
score=8 score=7 score=5
score=𝟗𝟗.𝟗𝟗𝟗𝟗
Compound
© 2016 Keisuke YANAGISAWA
Generalized Sum-3 (GS3)of fragment scores is adopted
12
Method
⋮Decomposition
O
CH3Cl
⋮
O Cl
O
F
OH
O
HN
⋮Evaluation
𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔=7
𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔=9.9
𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔=10.2
𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔=5
𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔𝒔=8
⋮
O
CH3Cl
⋮
O Cl
O
F
OH
O
HN
⋮
FragmentDocking
by Glide SP orGlide HTVS
© 2016 Keisuke YANAGISAWA
compounds fragments fragments compounds
13
Spresso is open-sourced under GPLv3 license(http://www.bi.cs.titech.ac.jp/spresso)
Acknowledgements
© 2016 Keisuke YANAGISAWA
