Különbözo szerkezetu porfirin származékok anti-HIV-1 aktivitásának modellezése

Translated title of the contribution: Modeling of anti-HIV-1 activity of a diverse set of porphyrin derivatives

Rozália Vanyúr, K. Heberger, J. Jakus

Research output: Contribution to journalArticle

Abstract

Various modeling methods {multiple linear regression (MLR), projection of latent structures (PLS) and artificial neural networks (ANN)} were compared in the modeling of anti-HIV-1 activity of porphyrin derivatives (see Figure 1, Table 1 and Table 2). The molecular structures were characterized by various topological and quantumchemical descriptors. Based on the three-dimensional structure of the geometry-optimized molecules, 87 theoretical descriptors were calculated by 3DNET program. These input descriptors were tested as independent variables and used for model building. All methods were suitable to build models to describe the studied biological activity (high R2 values in Table 4). The predictive abilities of the best fitting models were checked by leave-one-out (LOO) (Table 3 and Table 4), leave-n-out (LNO) cross-validation and also by external validation (Table 5). The models built by MLR could not predict the biological activities of compounds in the external validation set, while PLS models also gave bad results during the validation processes (LOO Q2=0.417). Only ANN models have shown good predictive ability in all validation steps, due to the inherent nonlinearity of the data sets (see figures). The degree of chemical bond rotational freedom descriptor (DF) was the most important variable, but could not predict the activity without other descriptors (Figure 2). The double bond equivalent descriptor (DBE) and WHIM descriptor weighted by atomic mass (KMASS), or energy of highest occupied molecular orbital (HOMO), or electrostatic total hydrogen bond acidity (ESTA) were also needed to predict anti-HIV-1 activity. The best prediction was achieved by a three-descriptor ANN model, model ANN_V (Figure 4). The external Q2 value of this model was 0.788, explaining 78.8% of the changes in the activity, Nevertheless, all of the ANN models were able to predict the studied activity with a Q2 > 0.6. Our ANN models were able to predict biological activities of a wide range of nontested tetrapyrrole molecules on the basis of their three-dimensional structures. Because of the low number of compounds in external validation set, we suggest to use the average value of biological activities calculated with the best four ANN models to predict anti-HIV-1 activity of similar compounds.

Original languageHungarian
Pages (from-to)71-75
Number of pages5
JournalMagyar Kemiai Folyoirat, Kemiai Kozlemenyek
Volume109-110
Issue number2
Publication statusPublished - Jun 2004

Fingerprint

Porphyrins
Derivatives
Neural networks
Bioactivity
Linear regression
Tetrapyrroles
Molecules
Chemical bonds
Molecular orbitals
Model structures
Acidity
Molecular structure
Electrostatics
Hydrogen bonds

ASJC Scopus subject areas

  • Chemistry(all)

Cite this

Különbözo szerkezetu porfirin származékok anti-HIV-1 aktivitásának modellezése. / Vanyúr, Rozália; Heberger, K.; Jakus, J.

In: Magyar Kemiai Folyoirat, Kemiai Kozlemenyek, Vol. 109-110, No. 2, 06.2004, p. 71-75.

Research output: Contribution to journalArticle

@article{bcc030b2210c4003b432ef5c0708d2db,
title = "K{\"u}l{\"o}nb{\"o}zo szerkezetu porfirin sz{\'a}rmaz{\'e}kok anti-HIV-1 aktivit{\'a}s{\'a}nak modellez{\'e}se",
abstract = "Various modeling methods {multiple linear regression (MLR), projection of latent structures (PLS) and artificial neural networks (ANN)} were compared in the modeling of anti-HIV-1 activity of porphyrin derivatives (see Figure 1, Table 1 and Table 2). The molecular structures were characterized by various topological and quantumchemical descriptors. Based on the three-dimensional structure of the geometry-optimized molecules, 87 theoretical descriptors were calculated by 3DNET program. These input descriptors were tested as independent variables and used for model building. All methods were suitable to build models to describe the studied biological activity (high R2 values in Table 4). The predictive abilities of the best fitting models were checked by leave-one-out (LOO) (Table 3 and Table 4), leave-n-out (LNO) cross-validation and also by external validation (Table 5). The models built by MLR could not predict the biological activities of compounds in the external validation set, while PLS models also gave bad results during the validation processes (LOO Q2=0.417). Only ANN models have shown good predictive ability in all validation steps, due to the inherent nonlinearity of the data sets (see figures). The degree of chemical bond rotational freedom descriptor (DF) was the most important variable, but could not predict the activity without other descriptors (Figure 2). The double bond equivalent descriptor (DBE) and WHIM descriptor weighted by atomic mass (KMASS), or energy of highest occupied molecular orbital (HOMO), or electrostatic total hydrogen bond acidity (ESTA) were also needed to predict anti-HIV-1 activity. The best prediction was achieved by a three-descriptor ANN model, model ANN_V (Figure 4). The external Q2 value of this model was 0.788, explaining 78.8{\%} of the changes in the activity, Nevertheless, all of the ANN models were able to predict the studied activity with a Q2 > 0.6. Our ANN models were able to predict biological activities of a wide range of nontested tetrapyrrole molecules on the basis of their three-dimensional structures. Because of the low number of compounds in external validation set, we suggest to use the average value of biological activities calculated with the best four ANN models to predict anti-HIV-1 activity of similar compounds.",
author = "Roz{\'a}lia Vany{\'u}r and K. Heberger and J. Jakus",
year = "2004",
month = "6",
language = "Hungarian",
volume = "109-110",
pages = "71--75",
journal = "Magyar Kemiai Folyoirat, Kemiai Kozlemenyek",
issn = "1418-9933",
publisher = "Magyar Kemikusok Egyesulete/Hungarian Chemical Society",
number = "2",

}

TY - JOUR

T1 - Különbözo szerkezetu porfirin származékok anti-HIV-1 aktivitásának modellezése

AU - Vanyúr, Rozália

AU - Heberger, K.

AU - Jakus, J.

PY - 2004/6

Y1 - 2004/6

N2 - Various modeling methods {multiple linear regression (MLR), projection of latent structures (PLS) and artificial neural networks (ANN)} were compared in the modeling of anti-HIV-1 activity of porphyrin derivatives (see Figure 1, Table 1 and Table 2). The molecular structures were characterized by various topological and quantumchemical descriptors. Based on the three-dimensional structure of the geometry-optimized molecules, 87 theoretical descriptors were calculated by 3DNET program. These input descriptors were tested as independent variables and used for model building. All methods were suitable to build models to describe the studied biological activity (high R2 values in Table 4). The predictive abilities of the best fitting models were checked by leave-one-out (LOO) (Table 3 and Table 4), leave-n-out (LNO) cross-validation and also by external validation (Table 5). The models built by MLR could not predict the biological activities of compounds in the external validation set, while PLS models also gave bad results during the validation processes (LOO Q2=0.417). Only ANN models have shown good predictive ability in all validation steps, due to the inherent nonlinearity of the data sets (see figures). The degree of chemical bond rotational freedom descriptor (DF) was the most important variable, but could not predict the activity without other descriptors (Figure 2). The double bond equivalent descriptor (DBE) and WHIM descriptor weighted by atomic mass (KMASS), or energy of highest occupied molecular orbital (HOMO), or electrostatic total hydrogen bond acidity (ESTA) were also needed to predict anti-HIV-1 activity. The best prediction was achieved by a three-descriptor ANN model, model ANN_V (Figure 4). The external Q2 value of this model was 0.788, explaining 78.8% of the changes in the activity, Nevertheless, all of the ANN models were able to predict the studied activity with a Q2 > 0.6. Our ANN models were able to predict biological activities of a wide range of nontested tetrapyrrole molecules on the basis of their three-dimensional structures. Because of the low number of compounds in external validation set, we suggest to use the average value of biological activities calculated with the best four ANN models to predict anti-HIV-1 activity of similar compounds.

AB - Various modeling methods {multiple linear regression (MLR), projection of latent structures (PLS) and artificial neural networks (ANN)} were compared in the modeling of anti-HIV-1 activity of porphyrin derivatives (see Figure 1, Table 1 and Table 2). The molecular structures were characterized by various topological and quantumchemical descriptors. Based on the three-dimensional structure of the geometry-optimized molecules, 87 theoretical descriptors were calculated by 3DNET program. These input descriptors were tested as independent variables and used for model building. All methods were suitable to build models to describe the studied biological activity (high R2 values in Table 4). The predictive abilities of the best fitting models were checked by leave-one-out (LOO) (Table 3 and Table 4), leave-n-out (LNO) cross-validation and also by external validation (Table 5). The models built by MLR could not predict the biological activities of compounds in the external validation set, while PLS models also gave bad results during the validation processes (LOO Q2=0.417). Only ANN models have shown good predictive ability in all validation steps, due to the inherent nonlinearity of the data sets (see figures). The degree of chemical bond rotational freedom descriptor (DF) was the most important variable, but could not predict the activity without other descriptors (Figure 2). The double bond equivalent descriptor (DBE) and WHIM descriptor weighted by atomic mass (KMASS), or energy of highest occupied molecular orbital (HOMO), or electrostatic total hydrogen bond acidity (ESTA) were also needed to predict anti-HIV-1 activity. The best prediction was achieved by a three-descriptor ANN model, model ANN_V (Figure 4). The external Q2 value of this model was 0.788, explaining 78.8% of the changes in the activity, Nevertheless, all of the ANN models were able to predict the studied activity with a Q2 > 0.6. Our ANN models were able to predict biological activities of a wide range of nontested tetrapyrrole molecules on the basis of their three-dimensional structures. Because of the low number of compounds in external validation set, we suggest to use the average value of biological activities calculated with the best four ANN models to predict anti-HIV-1 activity of similar compounds.

UR - http://www.scopus.com/inward/record.url?scp=33750295171&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33750295171&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:33750295171

VL - 109-110

SP - 71

EP - 75

JO - Magyar Kemiai Folyoirat, Kemiai Kozlemenyek

JF - Magyar Kemiai Folyoirat, Kemiai Kozlemenyek

SN - 1418-9933

IS - 2

ER -