benchmark

Description

Benchmarking of linear and non-linear approaches for QSPR studies of metal complexation with ionophoresIgor V. Tetko,1,2* Vitaly P. Solov'ev,3 Alexey V. Antonov,1 Xiaojun Yao,4 Jean Pierre Doucet,4 Botao Fan,4 Frank Hoonakker,5 Denis Fourches,5 Piere Jost,5 Nicolas Lachiche,5 and Alexandre Varnek,5 1- GSF - National Centre for Environment and Health, Institute for Bioinformatics(MIPS), 85764 Neuherberg, Germany2- Institute of Bioorganic & Petrochemistry, National Ukrainian Academy of Sciences, 02094, Kyiv, Ukraine, http://www.vcclab.org3- Institute of Physical Chemistry, Russian Academy of Sciences, Leninskiy prospect 31a, 119991 Moscow, Russia4- Université Paris 7-Denis Diderot, ITODYS-CNRS UMR 7086, 1, rue Guy de la Brosse, Paris 75005, France5- Laboratoire d'Infochimie, UMR 7551 CNRS, Université Louis Pasteur, 4, rue B. Pascal, Strasbourg 67000, FranceData SetsDescriptors+ 3+161 molecules 241 moleculeslogK (Ag ) logK (Eu ) 1 11O+ 3+ OH2log (Ag ) 112 molecules log (Eu ) 81 molecules 3 1'2 2Objectives OO 6 6' 4'4 2'H OOO HN C N N5 OH7 5' 7' 3'HOOH P COH NN OHOO Can we predict complexation constants NOOO S O OH OHOOSCl O OH using QSPR? OHO SEQUENCES AUGMENTEDATOMSOIIN IWhat are the best descriptors? ATOMS andBONDS(AB)HO N N HO OO=C-C-N; C-C-N; C-N; O=C-C; C=O; C-C C (-C) (-O) (=O)What are the best methods? HO N N ATOMS(A)NH2 N OH C (C) (O) (O) orO C C N; C C N; C N; O C C; C O; C C (Hy) C (C )(O )(O )sp2 sp3 sp3 sp2Do non-linear methods add ...

Informations

• Publié par: Vuin
• Nombre de lectures: 13
• Langue: English

Extrait

Benchmarking of linear and nonlinear approaches for QSPR studies of metal complexation with ionophores

Igor V. Tetko,1,2* Vitaly P. Solov'ev,3 Alexey V. Antonov,1 Xiaojun Yao,4 Jean Pierre Doucet,4 Botao Fan,4 Frank Hoonakker,5 Denis Fourches,5 Piere Jost,5 Nicolas Lachiche,5 and Alexandre Varnek,5

1 GSF  National Centre for Environment and Health, Institute for Bioinformatics(MIPS), 85764 Neuherberg, Germany 2 Institute of Bioorganic & Petrochemistry, National Ukrainian Academy of Sciences, 02094, Kyiv, Ukraine, http://www.vcclab.org 3 Institute of Physical Chemistry, Russian Academy of Sciences, Leninskiy prospect 31a, 119991 Moscow, Russia 4 Université Paris 7Denis Diderot, ITODYSCNRS UMR 7086, 1, rue Guy de la Brosse, Paris 75005, France 5 Laboratoire d'Infochimie, UMR 7551 CNRS, Université Louis Pasteur, 4, rue B. Pascal, Strasbourg 67000, France Objectives logK 1 (Ag + )  161 molec D ul a e t s a S l et K s 1 (Eu 3+ )  241 molecules Descriptors og O 1 + 2 OH lo 2 (Ag )  112 molecules lo 2 (Eu 3+ )  81 molecules 3 1' O O 6 6' 4' 4 2' O N H OH O C 7N 5 5 N ' 3' OH 7' OH HO  OH P C HO N N O xation constants O S O  O O Can we predict comple OH O HO O  N SPR? S O  using Q O O SEQUENCES AUGMENTED ATOMS Cä O  OH  OH II What are the best descriptors? N ATOMSandBONDS AB HO N  N HO O What are the best methods? O=CCN CCN CN O=CC C=O CC C C O =O  HO N N ATOMS A NH 2 N OH C C or O O Do nonlinear methods add some value? O C C N C C N C N O C C C O C C (Hy) C sp2 (C sp3 )(O sp3 )(O sp2 )   HO HO O BONDS B Can we compare results of different  O HO OH  HO    ==      =  =  C HO OH  O  O O O  methods in an objective way? O (C) Estate indices (D) Atomtype Estate indices and counts  atom index name value count values counts name index index O O no no O O  11.091 dO 44.35 4 1 1 SdO 1 dO(acid) 11.09 1 2 SdO(acid) 44.35 4 OH O  3 SdssC 4.08 42 dssC 1.02 1 HO HO  OH O  O 3 sOH 4 1 9.08 36.30 4 SsOH S  5 SsOH(acid) 1 36.30 4 9.083 sOH(acid) HO O OH  1 6 SssCH2 1.524 ssCH2 0.229 12 O OH  OH  O  O  4 6.57 7 SsssN 1 1.645 sssN 5 sssN(al) 1.64 1 8 SsssN(al) 6.57 4 O O  HO S 6 ssCH2 0.305 1 O O  O 7 ssCH2 0.305 1 O O O Analyzed approaches HO OH OH OH OH HO S A i s n s g o u c l i a a r t  i V v a e l  u N e e  u D r e a c l o N m et p w o o si r t k i  o ( n A ( S M N L N R ) A h / tt S p V :// D w ) w ht w tp .v :/ c /i c n l f a o b c . h o i r m g/l . a u b s / t a r s a n s n bg.fr/recherche/isida/ Traditional plot  Regression Error Curve Radial Basis Function Ne BF twork (R N) http://www.cs.waikato.ac.nz/~ml/weka eu k1 count _ Maximal Margin Linear Programming Method (MMLP) http://mips.gsf/proj/mdcs kNearest Neighbor Method (kNN) Support Vectors Machine http://www.csie.ntu.edu.tw/~cjlin/libsvm/

Data Analysis: double 5fold crossvalidation G e n e r a t i o n o f d e s c r i p to r s

Testing of Statistical Significance

aÄë Éêêçê D REC allows to compare results from several methods on one plot F r a g m e n t a l E  s t a t e c o u n t s E  s t a t e v a l u e s Statistical assesment of results 4 / 5 tr a in i n g s e t METHOD REC RMSE MAE R B F N  S V M ... K A S N N } rntecralsos liva ledoelesoitc ni ,nm aditno KolmogorovSmirnov (KS) .211.0NNre651.55 4.3331.0 AS cklab6.02SDV d N N  SVM 0.11 2.46 1.65  green Experimental versus predicted values for models o KNN 0.124 2.79 1.85  cyan 1 /5 t e s t s e t p r e d i c t i o n logK 1 (Ag+) and lo 2 (Eu3+). Despite apparent difference in   blue 07 1.98RBFN 0.132 3  . quality of both models, the outlying molecules in each model can be easil observed.   brown 2.22 3.89MMLP 0.142 AVERAGE 0.274 Sta tistic al e valu atio n  5.19 4.13  gray  BOOSTRAP: asnn > mmlp average p<0.001 Bootstrap significance test erag>av0.00ep<SORT 1 OBnknPA:1 BOOavpagerp<e000.ARTSs:P>mvlmm BOOSTRAP: weka > average p<0.001 BOOSTRAP: svd > average p<0.001 BOOSTRAP: mmlp > average p<0.001 Comparison of Methods  Comparison of Descriptors KS: svd != asnn 0.0081 1 1  svd != svm 0.0147 1 0 0  1 0 0 0 Éí E-ëí 1 a 9 0 2 9 0 2 îaäìÉë  svd != weka 0.0258 90 80 78 00 5 4 3 78 00 3 SMF  svd p<0.0001 != average 70 6 0 6 6 0 4 äì Ñ É ê ë a É Ö   ã î É a åíë -Eíaíë 60 KS: asnn != average p<0.0001 54 00  4  5 0 0 5 ëÉãíå50ÑêaÖSMF  KS: svm != mmlp 0.0258 40 3 0 7 03 02 E E- ë ë - íaíÉ Å    ìçaíÉíåí  svm != average p<0.0001 03 2 1 0 0 8 1 0 8 02 Åçìåí S: != p<0.0001 average K knn 0 0  p<0.0001KS: weka != average 01 1 2 3 4 5 6 1 2 3 4 5 6 7 8 0  p<0.0001KS: mmlp != average SMF fragments  Estate counts 1 2 3 4 5 6 7 8 9 01 1 0 0 1 Percentage of best models (y axis) calculated using Statistical analysis provides an objective comparison of different methods 9 0 2 corresponding descriptor system and all methods. 8 0 7 0 6 0 3 5 0 4 2 3 0 0 0 5 4 Conclusions 1 0 8 0 1 2 3 4 5 6 Models based on fragments (SMF, Estate counts) > Estate indices Estate indices  all descriptors Nonlinear approaches > multiple linear regression (MLRA) (p<0.05) But ensemble of several MLRA ≈ non-linear approaches Percentage of models (y axis) as a function of the number ofn t F o o p r   r e a a n c k h e  d d  a s t i a g  n s i et f  ic w a e n  t s  e m le o c d te el d s   n (  x dpercandcounnestdomsletseb atydateachperep.detceles)sixa Nosignificant differences in performance of nonlinear models ofmodelscontributedusingeachmethod.Calcula t t e ionswere SVM and ASNN provided largest number of "best" models perform ed using MLRA (1), RBF NN (2), kNN (3), MMLP (4), kNN was the fastest method ASNN (5), averaging of all ISIDA models (6), averaging of fi ve first ranked ISIDA m odels (7) and SVM (8).

Acknowledgement IVT was supported with Invited Professor position from Université Louis Pasteur. The part of this work has been performed in the framework of FrenchRussian collaborative project GDRE “SupraChem”.