PMID- 25191787
OWN - NLM
STAT- MEDLINE
DCOM- 20150626
LR  - 20151119
IS  - 1549-960X (Electronic)
IS  - 1549-9596 (Linking)
VI  - 54
IP  - 10
DP  - 2014 Oct 27
TI  - Prediction of compound potency changes in matched molecular pairs using support 
      vector regression.
PG  - 2654-63
LID - 10.1021/ci5003944 [doi]
AB  - Matched molecular pairs (MMPs) consist of pairs of compounds that are transformed 
      into each other by a substructure exchange. If MMPs are formed by compounds 
      sharing the same biological activity, they encode a potency change. If the 
      potency difference between MMP compounds is very small, the substructure exchange 
      (chemical transformation) encodes a bioisosteric replacement; if the difference 
      is very large, the transformation encodes an activity cliff. For a given compound 
      activity class, MMPs comprehensively capture existing structural relationships 
      and represent a spectrum of potency changes for structurally analogous compounds. 
      We have aimed to predict potency changes encoded by MMPs. This prediction task 
      principally differs from conventional quantitative structure-activity 
      relationship (QSAR) analysis. For the prediction of MMP-associated potency 
      changes, we introduce direction-dependent MMPs and combine MMP analysis with 
      support vector regression (SVR) modeling. Combinations of newly designed kernel 
      functions and fingerprint descriptors are explored. The resulting SVR models 
      yield accurate predictions of MMP-encoded potency changes for many different data 
      sets. Shared key structure context is found to contribute critically to 
      prediction accuracy. SVR models reach higher performance than random forest (RF) 
      and MMP-based averaging calculations carried out as controls. A comparison of SVR 
      with kernel ridge regression indicates that prediction accuracy has largely been 
      a consequence of kernel characteristics rather than SVR optimization details.
FAU - de la Vega de Leon, Antonio
AU  - de la Vega de Leon A
AD  - Department of Life Science Informatics, B-IT, LIMES Program Unit Chemical Biology 
      and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universitat , Dahlmannstr. 
      2, D-53113 Bonn, Germany.
FAU - Bajorath, Jurgen
AU  - Bajorath J
LA  - eng
PT  - Journal Article
DEP - 20140917
PL  - United States
TA  - J Chem Inf Model
JT  - Journal of chemical information and modeling
JID - 101230060
RN  - 0 (Ligands)
RN  - 0 (Receptors, Cell Surface)
RN  - 0 (Small Molecule Libraries)
SB  - IM
MH  - Algorithms
MH  - Humans
MH  - Isomerism
MH  - Ligands
MH  - Models, Molecular
MH  - Receptors, Cell Surface/agonists/antagonists & inhibitors/*chemistry
MH  - Regression Analysis
MH  - Small Molecule Libraries/*chemistry
MH  - *Software
MH  - Structure-Activity Relationship
MH  - *Support Vector Machine
MH  - Thermodynamics
EDAT- 2014/09/06 06:00
MHDA- 2015/06/27 06:00
CRDT- 2014/09/06 06:00
PHST- 2014/09/06 06:00 [entrez]
PHST- 2014/09/06 06:00 [pubmed]
PHST- 2015/06/27 06:00 [medline]
AID - 10.1021/ci5003944 [doi]
PST - ppublish
SO  - J Chem Inf Model. 2014 Oct 27;54(10):2654-63. doi: 10.1021/ci5003944. Epub 2014 
      Sep 17.