Skip Navigation


Carcinogenesis Advance Access originally published online on September 6, 2006
Carcinogenesis 2007 28(2):414-422; doi:10.1093/carcin/bgl159
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
28/2/414    most recent
bgl159v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (8)
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Manuguerra, M.
Right arrow Articles by Vineis, P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Manuguerra, M.
Right arrow Articles by Vineis, P.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© The Author 2006. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org

Multi-factor dimensionality reduction applied to a large prospective investigation on gene–gene and gene–environment interactions

M. Manuguerra1, G. Matullo1,2,*, F. Veglia1, H. Autrup3, A.M. Dunning4, S. Garte5, E. Gormally6, C. Malaveille6, S. Guarrera1, S. Polidoro1, F. Saletta1, M. Peluso7, L. Airoldi8, K. Overvad9, O. Raaschou-Nielsen10, F. Clavel-Chapelon11, J. Linseisen12, H. Boeing13, D. Trichopoulos14, A. Kalandidi14, D. Palli15, V. Krogh16, R. Tumino17, S. Panico18, H.B. Bueno-De-Mesquita19, P.H. Peeters20, E. Lund21, G. Pera22, C. Martinez23, P. Amiano24, A. Barricarte25, M.J. Tormo26, J.R. Quiros27, G. Berglund28, L. Janzon28, B. Jarvholm29, N.E. Day30, N.E. Allen31, R. Saracci6, R. Kaaks6, P. Ferrari6, E. Riboli32,33 and P. Vineis1,32,33

1 ISI Foundation, Torino Italy
2 Department of Genetics, Biology and Biochemistry University of Turin, Turin, Italy
3 Department of Environmental and Occupational Medicine, Aarhus Universitet Aarhus, Denmark
4 Department of Oncology, Strangeways Research Laboratory University of Cambridge, Cambridge, UK
5 Genetics Research Institute, Milan Italy
6 International Agency for Research on Cancer, Lyon France
7 Tuscany Cancer Institute, Cancer Risk Factor Branch, CSPO-Scientific Institute of Tuscany Florence, Italy
8 Istituto di Ricerche Farmacologiche Mario Negri, Milan Italy
9 Department of Clinical Epidemiology, Aalborg Hospital, Aarhus University Hospital Aalborg, Denmark
10 Institute of Cancer Epidemiology, Danish Cancer Society Copenhagen, Denmark
11 INSERM U521, Institut Gustave Roussy Villejuif, France
12 Division of Clinical Epidemiology, Deutsches Krebsforschungszentrum Heidelberg, Germany
13 German Institute of Human Nutrition, Potsdam-Rehbücke Germany
14 Department of Hygiene and Epidemiology, Medical School University of Athens, Greece
15 Molecular and Nutritional Epidemiology Unit, CSPO-Scientific Institute of Tuscany Florence, Italy
16 Department of Epidemiology, National Cancer Istitute Milan, Italy
17 Cancer Registry, Azienda Ospedaliera ‘Civile MP Arezzo’ Ragusa, Italy
18 Dipartimento di Medicina Clinica e Sperimentale, Università Federico II Naples, Italy
19 Centre for Nutrition and Health, National Institute for Public Health and the Environment Bilthoven, The Netherlands
20 Department of Cancer Epidemiology, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, The Netherlands
21 Institute of Community Medicine, University of Tromso Norway
22 Department of Epidemiology, Catalan Institute of Oncology Consejería de Sanidad y Servicios Sociales, Barcelona, Spain
23 Andalusian School of Public Health, Granada Spain
24 Department of Public Health of Guipuzkoa, San Sebastian Spain
25 Public Health Institute, Navarra Spain
26 Consejería de Sanidad y Consumo, Murcia Spain
27 Public Health and Health Planning Directorate, Asturias Spain
28 Malmö Diet and Cancer Study, Lund University Malmö, Sweden
29 Department of Public Health and Clinical Medicine, University of Umeå Sweden
30 MRC Dunn Human Nutrition Unit, Cambridge UK
31 Cancer Research UK Epidemiology Unit, University of Oxford UK
32 Imperial College London, London UK
33 University of Torino Italy

*To whom correspondence should be addressed. Section of Epidemiology, ISI Foundation (Institute for Scientific Interchange), Viale Settimio Severo, 65, 10133 Torino, Italy. Tel: +390116705601; Fax: +390112365601; Email: matullo{at}isiosf.isi.it


   Abstract
Top
 Abstract
Introduction
 Materials and methods
 Results
 Discussion
 References
 
It is becoming increasingly evident that single-locus effects cannot explain complex multifactorial human diseases like cancer. We applied the multi-factor dimensionality reduction (MDR) method to a large cohort study on gene–environment and gene–gene interactions. The study (case-control nested in the EPIC cohort) was established to investigate molecular changes and genetic susceptibility in relation to air pollution and environmental tobacco smoke (ETS) in non-smokers. We have analyzed 757 controls and 409 cases with bladder cancer (n = 124), lung cancer (n = 116) and myeloid leukemia (n = 169). Thirty-six gene variants (DNA repair and metabolic genes) and three environmental exposure variables (measures of air pollution and ETS at home and at work) were analyzed. Interactions were assessed by prediction error percentage and cross-validation consistency (CVC) frequency. For lung cancer, the best model was given by a significant gene–environment association between the base excision repair (BER) XRCC1-Arg399Gln polymorphism, the double-strand break repair (DSBR) BRCA2-Asn372His polymorphism and the exposure variable ‘distance from heavy traffic road’, an indirect and robust indicator of air pollution (mean prediction error of 26%, P < 0.001, mean CVC of 6.60, P = 0.02). For bladder cancer, we found a significant 4-loci association between the BER APE1-Asp148Glu polymorphism, the DSBR RAD52-3'-untranslated region (3'-UTR) polymorphism and the metabolic gene polymorphisms COMT-Val158Met and MTHFR-677C > T (mean prediction error of 22%, P < 0.001, mean CVC consistency of 7.40, P < 0.037). For leukemia, a 3-loci model including RAD52-2259C > T, MnSOD-Ala9Val and CYP1A1-Ile462Val had a minimum prediction error of 31% (P < 0.001) and a maximum CVC of 4.40 (P = 0.086). The MDR method seems promising, because it provides a limited number of statistically stable interactions; however, the biological interpretation remains to be understood.

Abbreviations: MDR, multi-factor dimensionality reduction; CVC, cross-validation consistency; DSBR, double-strand break repair; ETS, environmental tobacco smoke; FPRP, false positive report probability


   Introduction
Top
 Abstract
 Introduction
Materials and methods
 Results
 Discussion
 References
 
Environmental carcinogens contained in air pollution and environmental tobacco smoke (ETS) form mainly bulky DNA adducts and also generate interstrand cross-links and reactive oxygen species, which induce base damage, abasic sites, single- and double-strand breaks.

Many of the known environmental risk factors are metabolized in the organism to products that are either more carcinogenic or are detoxified. Genetic polymorphisms have been identified in many of these enzymes, and the biological consequence of some of such changes is an altered enzyme activity, which may influence the ratio between activation and deactivation, and thus the cancer risk (1). Enzymes involved in the biotransformation can be divided into phase 1 enzymes, involved in oxidative processes and phase 2 enzymes involved in the detoxification of the primary compounds or their metabolites. Genetic polymorphisms in both phase 1 and phase 2 metabolic enzymes have been shown to influence both cancer risk and the DNA adduct levels (2). We have demonstrated previously that people with high levels of adducts at the time they were healthy had a subsequent higher risk of developing cancer (3).

Unrepaired damage can result in apoptosis or may lead to unregulated cell growth and cancer (4). Alternatively, the damage can be repaired at the DNA level enabling the cell to replicate as planned. Because of the importance of maintaining genomic integrity in the general and specialized functions of cells as well as in the prevention of carcinogenesis, genes coding for DNA repair molecules have been proposed as candidate cancer-susceptibility genes (5,6).

Studies to date indicate that variation in DNA repair and metabolic genes may influence cancer susceptibility (7,8); however, results are not fully consistent, and other potentially important polymorphisms in these and other genes have not been explored yet.

Based on the extensive reviews of the literature (8–11) several polymorphisms of DNA repair, metabolic and other genes potentially associated with lung cancer, bladder cancer and myeloid leukemia have been analyzed in the context of the GEN-AIR project, a case-control study nested in the EPIC investigation. The aim of the present paper is to apply the model-free multi-factor dimensionality reduction (MDR) method (12) to a large study whose primary goals were to investigate the role of air pollution and ETS in increasing the risk of cancer, and to investigate gene–environment and gene–gene interactions.

In fact, it is becoming increasingly evident that single-locus effects cannot explain complex multifactorial human diseases like cancer. Recent reviews and meta-analyses have demonstrated the difficulties associated with replicating main effect results. Thus, when the single polymorphism effect is not present alone or is not strong enough, the identification and characterization of susceptibility genes for cancer risk require the understanding of gene–gene interactions.


   Materials and methods
Top
 Abstract
 Introduction
 Materials and methods
Results
 Discussion
 References
 
Subjects
European Prospective Investigation into Cancer and Nutrition (EPIC) is a multicenter European study, coordinated by the International Agency for Research on Cancer (Lyon), in which more than 500 000 healthy volunteers were recruited in 10 European countries (France, Denmark, Germany, Greece, Italy, The Netherlands, Norway, Spain, Sweden and UK) corresponding to 23 recruitment centers. The cohort includes subjects of both genders, mostly in the age range 35–74 at recruitment. Recruitment took place between 1993 and 1998. Detailed dietary and lifestyle histories collected mainly through self-administered questionnaires, plus a 24 h dietary recall through person-to-person interview (in a 10% sample), anthropological measurements and a 30–40 ml blood sample are available. All questionnaire information is available in a computerized format. Signed informed consent forms were collected from all participants (except a sub-group of the Oxford cohort who gave consent on postal questionnaires).

Gen-Air is a case-control study nested within the EPIC cohort, aiming at studying the relationship between some types of cancer and air pollution or ETS. Cases are subjects with bladder, lung, oral, pharyngeal or laryngeal cancers and leukemia, all newly diagnosed after recruitment. Also deaths from respiratory diseases (COPD and emphysema) were identified and included. Only non-smokers or ex-smokers since at least 10 years have been included in Gen-Air. We have matched three controls per case for exposure assessment and the analysis of questionnaire data, and two controls per case for laboratory analyses. Matching criteria were gender, age (±5 years), smoking status (never or former smoker), country of recruitment, and follow-up time. Mean follow-up was 89 months (minimum 51 and maximum 123).

Gen-Air has been approved by the Ethical Committee of the International Agency for Research on Cancer, and by the local Ethical Committees of the 23 centres.

We have identified 4051 subjects (1074 cases and 2977 controls) in EPIC who met the Gen-Air protocol criteria. The distribution of cases by cancer site or cause of death was as follows: bladder cancer 227, lung cancer 271, oral, pharyngeal cancer 73, laryngeal cancer 58, leukemias 311, deaths from respiratory diseases 134. Of these subjects, 2410 had blood samples (846 cases and 1564 controls). The Malmo center has decided not to allow the use of their blood samples and the Umea center did not allow genetic analyses, but they participate in the rest of the project.

In the present study, we have analyzed 409 cases of bladder cancer (n = 124), lung cancer (n = 116), leukemia (n = 169) and 757 controls with available blood samples and successful DNA extraction and genotype analysis. The remaining endpoints were not analyzed as the methods used in this paper require a large number of subjects. Our approach did not keep the matching between cases and controls.

In addition to several gene variants (see below) we also included in the analyses three exposure variables, air pollution (as indirectly measured by the distance from a heavy traffic road) and exposure to ETS at work and at home. For details on these exposures and the ways they were measured see our previous publications (13,14).

DNA extraction and genotype analysis
The choice of the relevant polymorphisms (Table I) has been made on the basis of an extensive review of the literature (8–11) and encompasses genes involved directly or indirectly in DNA repair and metabolic genes.


View this table:
[in this window]
[in a new window]

  		Table I Gene variants included in the MDR model

 
DNA was extracted from 200–300 µl of buffy coats in Genova and Florence laboratories. DNA was isolated and purified as described in Peluso et al. (15).

Genotyping has been performed with Light Cycler at IARC by Dr Malaveille (genes: MnSOD, COMT, MPO, SULT-1A1), with Taqman at the ISI Foundation in Torino by Dr Matullo (XPD, XRCC1, PCNA, ERCC1, MGMT, OGG1) and at the University of Cambridge by Dr Dunning (BRCA1 and 2, NBS1, RAD51-52, XRCC2, LIG4, TP53), with a PCR–RFLP based assay at the University of Aarhus by Dr Autrup (GSTM1, GSTM3, GSTT1, GSTP, NQO1) and at the Genetics Research Institute in Milan by Dr Garte (NAT2, CYP1A1, CYP1B1, MTHFR).

MDR analysis
The common analytical tool to analyse gene–gene interactions under a multiplicative model is logistic regression analysis. Unfortunately, parametric and model-based statistical methods require specific hypotheses to be tested; in high-dimensional analysis such as in the investigation of all potential interactions, the number of hypotheses is inflated to a great extent. To address this issue it has been suggested that model-free, data-based exploratory methods are more flexible and powerful.

A MDR reduction method with related open-source software has been developed by Ritchie et al. (12) to detect and characterize high-order gene–gene and gene–environment interactions in studies with relatively small sample sizes.

The goal of MDR is to find the main factor and the combinations of 2, ... , N factors that are more frequently associated with case than with control status (adjusted for the ratio between them). The MDR method was applied as described previously (12,16–22). Briefly, to search for the best n-loci model (with n = 1, ... , N), the dataset is randomly divided into 10 equal parts. A training set of 9/10 of the data are used to search for the best model, i.e. to classify each genotype combination as a high-risk or a low-risk pattern, depending on the number of cases and controls that present that combination. In Figure 1 each box represents a particular combination of genotypes for a 4-loci model (Figure 1A) and a 3-loci model (Figure 1B). If the box is dark, the combination is a high-risk pattern, i.e. we observe that the particular combination of genotypes for these polymorphisms is more frequently associated with the case status than with the control status. In Figure 1B the box representing the combination of the w/w genotype (w = wild-type, m = mutant) in CYP1A1-Ile462Val polymorphism, the w/m genotype in RAD52-2259C > T and the m/m genotype in MnSOD-Ala9Val is labelled as high-risk, although the number of cases is only 13 and that of controls is 22. The reason is that their ratio (13/22 = 0.59) is greater than the ratio of the number of cases to the number of controls for the leukemia dataset (169/305 = 0.55). The remaining 1/10 of the data are the testing set, used to control the goodness-of-fit of the model. This procedure is repeated 10 times, in order to use all the possible testing sets.


Figure 1
View larger version (25K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Fig. 1 Genotype combinations labelled as high-risk and low-risk for bladder cancer and leukemia. In each box the left and the right bars represent the number of cases and controls for each combination of genotypes. If the ratio of cases to controls exceeds a threshold (the ratio of total number of cases to total number of controls in the present analysis), the combination is labelled as high risk (dark boxes). (A) Bladder cancer. (B) Leukemia.

 
For each n-loci model the MDR method gives two scores, a mean prediction error percentage and the cross-validation consistency (CVC) frequency. The former is the proportion of subjects for whom an incorrect class prediction was made, while the latter is the number of times a particular combination of loci (model) is identified in each possible testing set. The best model is that with lower prediction error and maximum CVC. We repeated the complete analysis 10 times, using different random seeds to reduce the probability of biased results due to the chance divisions of the data in training and testing sets.

To evaluate the magnitude of the prediction error and the CVC, we permutated the status of cases and controls in the dataset and repeated the analysis 1000 times, obtaining for each n-loci model the prediction error and the CVC distributions under the null hypothesis of no association. Comparing the results and these distributions we obtained the P-values associated with each prediction error and CVC.

For each model, we computed the associated odds ratio (Table II). The reference group (unexposed) is formed by subjects presenting the combination of genotypes labelled as low-risk, while the exposed group is that with subjects that present the combination labelled as high-risk.


View this table:
[in this window]
[in a new window]

  		Table II Results of MDR for lung and bladder cancers and leukemia

 
Limitations of MDR
Although MDR overcomes some of the limitations of the generalized linear model, there are also some other limitations to consider (12,19). First, MDR is computationally intensive for high-order models and would require genetic search algorithms to find solutions in a limited time. Second, MDR results are difficult to interpret. If a strong main effect is present, it is not simple to understand the importance of the other polymorphisms in the model. More generally, there is no information on the relations between the polymorphisms involved in each model; e.g. if the best model is a four-way interaction, it is not clear if it is the sum of two separate two-way interactions or a two-way interaction and two main effects, etc. Third, high dimensionality and a small dataset may lead to many multi-factor cells with either a small number of subjects or no subjects. It is not clear how the model sensitivity changes with the dimension of the dataset. For example, a source of noise is given by missing values. The proportion of missing values can vary among cases and controls and thus affect the results. In our dataset, the mean differences in percentage of missing values between cases and controls are <3%. Finally, MDR assumes that there is no genetic heterogeneity. If a group of cases are explained by a combination of loci different from the one that explains another group of cases, the MDR method can fail in finding the correct models.

Hypotheses tested
The main advantage of using a data-based statistical method like MDR is the ability to analyse a dataset without the need to hypothesize an a priori model (model free analysis). At the same time, it can be useful to test for some hypotheses, such as specifying the recessive or dominant model of inheritance. To simplify the analysis, we have searched for scenarios in which all the gene variants show a recessive or a dominant behaviour. In these analyses, we have assigned the genotypes with the highest frequency as wild-type for each polymorphism and used them as the reference group.

False positive report probability (FPRP)
Reports of associations between polymorphisms and complex diseases are greatly affected by the risk of being false positives. To estimate the magnitude of this probability we used the method described by Wacholder (23) and recently applied in some epidemiological studies (24–26). To compute the FPRP, we used the odds ratios from MDR models, in which we consider the given classification of high risk and low-risk genotype combinations, the P-values and the power to detect ORs of 2 and 4. The P-values are calculated from the comparison of the proportion of correct predictions of the model and the distribution of randomized datasets.

As there are many factors influencing the prior probability that the association between a genetic variant and a disease is real (27), the FPRP is commonly computed for prior probabilities ranging in a wide interval [1.E-5 to 0.25 in Wacholder (23), 0.001 to 0.50 in Hung (24) and Matullo (25,26)]. Considering the lack of information on the interactions between genes and environmental variables, in our study we have considered a wider interval for prior probabilities, with bottom and top values lower than those used in other studies, i.e. from 1.E-6 to 0.10.

Validation and interpretation of the results using information gain and interaction graphs
The results obtained with the MDR method were validated using an approach based on measures of information and entropy. Jakulin and Bratko (28) have provided a generalization of McGill's interaction information (29) to evaluate the information gain (IG) related to a class variable (e.g. case-control status) by merging several attributes together, over the information provided by the attributes independently.

Let H(X) be the Shannon entropy of X (29). In the case with two attributes A, B and a class label C, the IG of A, B and C can be written as IG(A;B;C) = I(A;B|C) – I(A;B), where I(A;B|C) = H(A|C) + H(B|C) – H(A,B|C) and I(A;B) = H(A) + H(B) – H(A,B). I(A;B) is the mutual information between A and B, while I(A;B|C) is the conditional mutual information and measures the relationship between A and B in the context of C.

Generalizing from formulas in (29), Jakulin and Bratko define the k-way interaction information for k = 3, 4 to an arbitrary k as:

Formula

where A={X1, X2, ... , Xn} is a set of n attributes (e.g. the variables measured in the study) and S {subseteq} A is a subset of k attributes.

As the softwares used to implement these formulas do not permit missing values, their values have been randomly imputed 10 times respecting the genotype proportions in the data. The results reported are the mean values obtained from the analyses performed using each of the 10 datasets.

Computing the interaction information for all the possible combinations of k attributes chosen in a set of 40 variables (exposures and SNPs) is a time consuming process for k > 3 at the current speed of computers. To address this issue we used a two-step procedure following the indications given by Moore et al. (30). First a subset of more informative attributes to analyse is selected. To do this, the IG is estimated for each individual attribute and each pairwise combination of attributes. Pairs of attributes are sorted and those with the highest IG, or percentage of entropy removed, are selected for further consideration. The selection of the cut-off to exclude the attributes is made considering the distribution of the IG values for each pairwise combination. Usually, these distributions have a Gaussian-like shape, with only few interactions standing out from the crowd. In the second step, the IG is estimated for each attribute and each combination of two, three and four attributes using only the selected subset.

Measures of IG for main and interaction effects are useful to plot interaction graphs that can give indications to interpret the relationship between attributes. In the interaction graphs nodes and connections indicate the percentage of entropy in case-control status removed by each variable (main effect) and by each pairwise combination of attributes (interaction effect). A positive entropy indicates interaction while a negative entropy indicates redundancy.

Software
All MDR computations to find and test the models have been performed using the open-source java software MDR v.1.0.0rc1. FPRP results are calculated with the Excel spreadsheet distributed at http://jncicancerspectrum.oupjournals.org/jnci/content/vol96/issue6 (23). The formulas from Jakulin and Bratko have been implemented in R, v.2.2.1 (R Foundation for Statistical Computing, Vienna, Austria). The Orange software package (31) has been used to test the routines written in R and to build the interaction graphs.


   Results
Top
 Abstract
 Introduction
 Materials and methods
 Results
Discussion
 References
 
Table II shows for each cancer site and each number of loci/exposure variables evaluated (up to 4), the average CVC, the average prediction error and the associated P-values obtained from the randomized analysis of lung and bladder cancer, and leukemia datasets.

For lung cancer, the best result (model free analysis) is provided by a significant gene–environment interaction between XRCC1-Arg399Gln, BRCA2-Asn372His polymorphisms and the exposure variable ‘heavy traffic road’ (mean prediction error of 26%, P < 0.001, and mean CVC of 6.60, P = 0.02). The exposure variable represents distance from a heavy traffic road and is an indirect and robust indicator of air pollution. Moreover, a 4-loci interaction under a recessive model (ERCC1-Asn118Asn, BRCA2-Asn372His, RAD52-2259C > T and NQO1-Pro187Ser polymorphisms) showed also significant results (Table II, P < 0.001 for mean prediction error and P = 0.096 for CVC). The 4-loci interaction was regarded as a better model than the encapsulated 3-loci model (that showed comparable reliability for mean prediction error and CVC) because had lower FPRP values (Table III).


View this table:
[in this window]
[in a new window]

  		Table III False positive report probabilities and odds-ratios for the best models

 
For bladder cancer, we found a significant 4-loci interaction between APE1-Asp148Glu, RAD52-2259C > T and the metabolic gene polymorphisms COMT-Val158Met and MTHFR-677C > T (mean prediction error of 22%, P < 0.001 and mean CVC of 7.40, P = 0.013–0.037).

For leukemia, a 3-loci interaction including RAD52-2259C > T, MnSOD-Ala9Val and CYP1A1-Ile462Val, had a minimum prediction error of 31% (P < 0.001) and a maximum CVC of 4.40 (P = 0.086). For both bladder cancer and leukemia, the recessive and dominant models did not yeld any significant results.

Table III reports the FPRP values calculated using the statistical power to detect ORs of 2 and 4 with {alpha} level equal to the observed P value. Results show an excellent reliability for the bladder 4-loci and leukemia 3-loci models, even assuming very low prior probabilities (<0.001). Figure 1A and B show the detailed models.

On the other hand, the two interaction models found for lung cancer seem to indicate false positive reports (as shown in Table III for OR = 2), since unrealistic prior probabilities are necessary to obtain FPRP values below 0.2.

In the two-step validation method, all the attributes involved in the MDR models for lung cancer, bladder cancer and leukemia were present in the more informative subset of variables selected by the IG method when used for main and interaction effects. In the second step, the lung cancer and leukemia models did not obtain high IG values, i.e. there were many models with other attributes (more than 10) reporting higher IGs. The bladder cancer model, on the other side, resulted to be the best model also for the IG method. The percentage of entropy removed by each variable and by each pairwise combination of attributes is shown in Figure 2. The values are all positive, indicating a positive interaction between all the polymorphisms included the model.


Figure 2
View larger version (11K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Fig. 2 Orange canvas interaction graph. Nodes and connections indicate the percentage of entropy in case-control status removed by each variable (main effect) and by each pairwise combination of attributes (interaction effect). The values are all positive, indicating a positive interaction between all the polymorphisms of the model.

 

   Discussion
Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
References
 
In the present study, we investigated the possible combined contribution of genetic and environmental factors to determining complex cancer risk patterns through gene–gene and gene–environment interactions. We applied a robust and validated method to reduce the complexity of multi-factor analysis in a large prospective investigation on air pollution, ETS and cancer. The results were validated using a generalization to many attributes of the mutual information method, a well-established approach to evaluate interactions between two variables. The attributes selected in the lung cancer and leukemia models were factors in the subset of the more informative variables, while the MDR best model was confirmed by the IG method as the best 4-way interaction for bladder cancer (Figure 2). The patterns we have identified need further biological investigation for interpretation beyond their statistical significance. The biological plausibility of the interpretation is in fact a test of the viability of multidimension reduction methods.

The first consideration about the emerging patterns is that DNA repair genotypes seem to play an important role in increasing cancer risk. Amongst the genes that seem to be involved in the risk of lung and bladder cancer and of leukemia, RAD52 emerges repeatedly in our analysis, followed by XRCC1, BRCA2 and APE1. Although we investigated a limited number of DNA repair genes/polymorphisms, it seems that impaired base excision repair (APE1 and XRCC1) and double-strand break repair (RAD52 and BRCA2) could be common pathways leading to different forms of cancer. Unfortunately the literature on DNA repair gene variants and cancer risk is still scanty and unclear, reporting often contrasting results (8,32,33). The involvement of proximity to a heavy traffic road is the only environmental factor that interacts significantly with gene variants, and it is coherently associated with lung cancer (13). Distance from a heavy traffic road is an indirect and robust indicator of air pollution. In bladder cancer, APE1 and RAD52 DNA repair gene variants strongly interacted with some metabolic genes, namely COMT and MTHFR, in increasing the risk. APE1 is the major 5' AP endonuclease of mammalian cells that is responsible for the incision of baseless lesions in DNA during BER. Although the APE1-Asp148Glu variant allele seems to have a small effect on AP endonuclease and DNA-binding activities (34) an increased risk of chromosomal aberration has been observed for this polymorphism (35). RAD52 is involved in DSBR through homologous recombination and facilitates strand exchange reaction catalysed by RAD51. Although the RAD52 gene could be a good candidate due to the high number of aberrations and allelic losses in different chromosomes observed in bladder cancer cells, for the RAD52-2259C > T polymorphism insufficient information has been published for any conclusion about its functional consequences to be drawn.

COMT catalyses the methylation of various endobiotic and xenobiotic substances preventing quinone formation and redox cycling, and therefore might protect DNA from oxidative damage (1). A G to A transition, which results in amino acid change from valine to methionine at codon 108, leads to a lower COMT activity in a co-dominant manner (36). In a previous study we have analyzed the tumour biopsies of 45 patients with bladder cancer for p53 mutations by direct sequencing. All p53 mutations occurred in subjects with the COMT variant allele (P = 0.03) (37). Although, a recent study on bladder cancer did not find an association with the COMT variant (38), a role of this enzyme in bladder cancer is likely, as supported also by the concomitant use of a soluble isoform of COMT, calreticulin and gamma-synuclein as urinary markers to improve the diagnosis of bladder cancer (39). MTHFR is an enzyme that plays a key role in the metabolism of folate and in DNA synthesis, and common polymorphisms seem to affect its functional properties (40). The MTHFR-677 variant genotypes (CT or TT) are common in the population. The studies on MTHFR and bladder cancer risk are inconsistent, with negative results in three studies (41–43), while a complex pattern was apparent in another investigation, in which individuals carrying the variant genotypes (CT or TT) and reporting a low folate intake were at a significantly 3.5-fold increased risk of bladder cancer (95% CI: 1.6–6.5) (44). In contrast, individuals carrying a variant genotype and reporting a high folate intake were at a 1.4-fold increased risk (95% CI: 0.7–2.7), and wild-type homozygotes reporting a low folate intake were at only 1.6-fold increased risk (95% CI: 0.8–3.0). The interaction between genetic polymorphisms and folate intake was significant on the multiplicative scale (P = 0.01) (44). The role of folate in inducing bladder cancer is biologically plausible, as suggested by the protective effect of fruit and vegetables on bladder cancer (45), but the hypothesis of an involvement of MTHFR is still uncertain.

Less clear is the lack of involvement (even in terms of interactions) of metabolic gene polymorphisms, such as NAT-2 and GSTM1, that are established genetic risk factors for bladder cancer as shown in recent large case-control studies (46). However, this difference may be due to the fact that our population includes only non-smokers. Similar considerations can also apply to the results on lung cancer and leukemia. In particular, for lung cancer we have not confirmed the involvement of XRCC1-Arg194Trp and of hOGG1-Ser326Cys polymorphisms, as recently suggested (24,32). However, we have identified a possible weak interaction of NQO1-Pro187Ser polymorphism in combination with ERCC1-Asn118Asn, BRCA2-Asn372His, RAD52-2259C > T polymorphisms; a positive association with lung cancer has been recently reported for NQO1-Pro187Ser polymorphism (47).

For leukemia, the involvement of a DSBR gene, i.e. RAD52, is plausible as suggested in different studies, since chromosomal translocations are very common genetic abnormalities in this disease (48,49), with evidence of loss of heterozygosity for a chromosomal segment including the RAD52 locus in T cell prolymphocytic leukaemia (50). Less clear is the involvement of a free radical scavenger such as MnSOD, although there are several studies reporting on down-regulation of this enzyme after treatment of leukaemic cells with different chemotherapeutic agents (51,52). However, no study investigated the relationship between leukaemia and MnSOD polymorphisms. On the other hand, CYP1A1 polymorphisms have been largely investigated in different types of leukaemias, showing positive associations in childhood leukaemia (53), and acute adult lymphoblastic leukaemia (54,55). A previously described interaction with the NAT-2 and GSTM1 genes (56) has not been confirmed in our study.

In summary, the method of multifactor-dimensionality reduction, even with the limitations previously discussed, seems promising. In fact, it provides a limited number of statistically stable associations among genes and between genes and the environment that have been validated by an alternative and well documented method. However, the lack of clear biological background knowledge hampers a clear interpretation of the findings.


   Acknowledgments
 
The authors wish to thank to Angeline Andrew and Margaret Karagas for their useful comments. This paper was made possible by European Community grants to PV (5th Framework Programme, GENAIR investigation, grant QLK4-CT-1999-00927; 6th Framework Programme, ECNIS investigation, grant 513943) and a grant of the Compagnia di San Paolo to the ISI Foundation. All authors are independent from funders. Also, the work described in the paper was carried out with the financial support of: Europe Against Cancer Program of the European Commission (SANCO); Deutsche Krebshilfe; Deutsches Krebsforschungszentrum; German Federal Ministry of Education and Research; Danish Cancer Society; Health Research Fund (FIS) of the Spanish Ministry of Health; Spanish Regional Governments of Andalucia, Asturia, Basque Country, Murcia and Navarra; ISCIII, Red de Centros RCESP, C03/09, Spain; Cancer Research UK; Medical Research Council, United Kingdom; Stroke Association, United Kingdom; British Heart Foundation; Department of Health, United Kingdom; Food Standards Agency, United Kingdom; Wellcome Trust, United Kingdom; Greek Ministry of Health; Greek Ministry of Education; Italian Association for Research on Cancer (AIRC); Italian National Research Council; Dutch Ministry of Public Health, Welfare and Sports; World Cancer Research Fund; Swedish Cancer Society; Swedish Scientific Council; Regional Government of Skåne, Sweden; Norwegian Cancer Society; Research Council of Norway.

Conflict of Interest Statement: None declared.


   References
Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 

  1. Autrup H. (2000) Genetic polymorphisms in human xenobiotica metabolizing enzymes as susceptibility factors in toxic response. Mutat Res. 464:65–76.[ISI][Medline]
  2. Autrup H. (2004) Gene-environment interaction in environmental carcinogenesis. In Nicolopou-Stamati H. (Ed.), et al. Cancer as an Environmental DiseaseKluwer Academic Publishers.
  3. Peluso M., Munnia A., Hoek G., Krzyzanowski M., Veglia F., Airoldi L., Autrup H., Dunning A., Garte S., Hainaut P., et al. (2005) DNA adducts and lung cancer risk: a prospective study. Cancer Res. 65:8042–8048.[Abstract/Free Full Text]
  4. Vispe S., Yung T.M., Ritchot J., Serizawa H., Satoh M.S. (2000) A cellular defense pathway regulating transcription through poly (ADP-ribosyl)ation in response to DNA damage. Proc. Natl Acad. Sci. USA 97:9886–9891.[Abstract/Free Full Text]
  5. Squire J.A., Whitmore G.F., Phillips R.A. (1998) Genetic basis of cancer. In Tannock I.F., Tannock I., Hill R.P. (Eds.). The Basic Science of Oncology 3rd edn McGraw-Hill Press.
  6. Knudson A.G. Jr. (1989) The genetic predisposition to cancer. Birth Defects Orig. Artic. Ser. 25:15–27.[Medline]
  7. Friedberg E.C., Walker G.C., Siede W. (1995) DNA Repair and Mutagenesis(ASM Press, Washington, DC).
  8. Goode E.L., Ulrich C.M., Potter J.D. (2002) Polymorphisms in DNA repair genes and associations with cancer risk. Cancer Epidemiol. Biomarkers Prev. 11:1513–30.[Abstract/Free Full Text]
  9. Au W.W., Navasumrit P., Ruchirawat M. (2004) Use of biomarkers to characterize functions of polymorphic DNA repair genotypes. Int. J. Hyg. Environ. Health 207:301–313.[CrossRef][ISI][Medline]
  10. Berwick M., Matullo G., Vineis P. (2002) Studies of DNA repair and human cancer: an update. In Wilson S.H. and Suk W.A. (Eds.). Biomarkers of Environmentally Associated Disease: Technologies, Concepts and Perspectives(Lewis publishers, CRC Press LLC, NY) pp. 83–107.
  11. Berwick M. and Vineis P. (2000) Markers of DNA repair and susceptibility to cancer in humans: an epidemiologic review. J. Natl Cancer Inst. 92:874–97.[Abstract/Free Full Text]
  12. Ritchie M.D., Hahn L.W., Roodi N., Bailey L.R., Dupont W.D., Parl F.F., Moore J.H. (2001) Multifactor-dimensionality reduction reveals high-order interactions among estrogen-metabolism genes in sporadic breast cancer. Am. J. Hum. Genet. 69:138–147.[CrossRef][ISI][Medline]
  13. Vineis P., Hoek G., Krzyzanowski M., Vigna-Taglianti F., Veglia F., Airoldi L., Autrup H., Dunning A., Garte S., Hainaut P., et al. (2006) Air pollution and risk of lung cancer in a prospective study in Europe. Int. J. Cancer.
  14. Vineis P., Airoldi L., Veglia P., Olgiati L., Pastorelli R., Autrup H., Dunning A., Garte S., Gormally E., Hainaut P., et al. (2005) Environmental tobacco smoke and risk of respiratory cancer and chronic obstructive pulmonary disease in former smokers and never smokers in the EPIC prospective study. BMJ 330:277.[Abstract/Free Full Text]
  15. Peluso M., Hainaut P., Airoldi L., Autrup H., Dunning A., Garte S., Gormally E., Malaveille C., Matullo G., Munnia A., et al. (2005) Methodology of laboratory measurements in prospective studies on gene-environment interactions: the experience of GenAir. Mutat Res. 574:92–104.[ISI][Medline]
  16. Hahn L.W., Ritchie M.D., Moore J.H. (2003) Multifactor dimensionality reduction software for detecting gene–gene and gene–environment interactions. Bioinformatics 19:376–382.[Abstract/Free Full Text]
  17. Ritchie M.D., Hahn L.W., Moore J.H. (2003) Power of multifactor dimensionality reduction for detecting gene–gene interactions in the presence of genotyping error, missing data, phenocopy, and genetic heterogeneity. Genet. Epidemiol. 24:150–157.[CrossRef][ISI][Medline]
  18. Cho Y.M., Ritchie M.D., Moore J.H., Park J.Y., Lee K.U., Shin H.D., Lee H.K., Park K.S. (2004) Multifactor-dimensionality reduction shows a two-locus interaction associated with Type 2 diabetes mellitus. Diabetologia 47:549–554.[CrossRef][ISI][Medline]
  19. Coffey C.S., Hebert P.R., Ritchie M.D., Krumholz H.M., Gaziano J.M., Ridker P.M., Brown N.J., Vaughan D.E., Moore J.H. (2004) An application of conditional logistic regression and multifactor dimensionality reduction for detecting gene–gene interactions on risk of myocardial infarction: the importance of model validation. BMC Bioinformatics 5:49.[CrossRef][Medline]
  20. Williams S.M., Ritchie M.D., Phillips J.A. 3rd, Dawson E., Prince M., Dzhura E., Willis A., Semenya A., Summar M., White B.C., et al. (2004) Multilocus analysis of hypertension: a hierarchical approach. Hum. Hered. 57:28–38.[CrossRef][ISI][Medline]
  21. Ritchie M.D. and Motsinger A.A. (2005) Multifactor dimensionality reduction for detecting gene–gene and gene–environment interactions in pharmacogenomics studies. Pharmacogenomics 6:823–834.[CrossRef][ISI][Medline]
  22. Martin E.R., Ritchie M.D., Hahn L., Kang S., Moore J.H. (2006) A novel method to identify gene–gene effects in nuclear families: the MDR-PDT. Genet. Epidemiol. 30:111–123.[CrossRef][ISI][Medline]
  23. Wacholder S., Chanock S., Garcia-Closas M., El Ghormli L., Rothman N. (2004) Assessing the probability that a positive report is false: an approach for molecular epidemiology studies. J. Natl Cancer Inst. 96:434–442.[Abstract/Free Full Text]
  24. Hung R.J., Brennan P., Canzian F., Szeszenia-Dabrowska N., Zaridze D., Lissowska J., Rudnai P., Fabianova E., Mates D., Foretova L., et al. (2005) Large-scale investigation of base excision repair genetic polymorphisms and lung cancer risk in a multicenter study. J. Natl Cancer Inst. 97:567–576.[Abstract/Free Full Text]
  25. Matullo G., Dunning A.M., Guarrera S., Baynes C., Polidoro S., Garte S., Autrup H., Malaveille C., Peluso M., Airoldi L., et al. (2005) DNA repair polymorphisms and cancer risk in non-smokers in a cohort study. Carcinogenesis.
  26. Matullo G., Guarrera S., Sacerdote C., Polidoro S., Davico L., Gamberini S., Karagas M., Casetta G., Rolle L., Piazza A., et al. (2005) Polymorphisms/haplotypes in DNA repair genes and smoking: a bladder cancer case-control study. Cancer Epidemiol. Biomarkers Prev. 14:2569–2578.[Abstract/Free Full Text]
  27. Matullo G., Berwick M., Vineis P. (2005) Gene–environment interactions: how many false positives? J. Natl Cancer Inst. 97:550–551.[Free Full Text]
  28. Jakulin A. and Bratko I. (2003) Analyzing attribute interactions. Lect. Notes Artiff. Intell. 2838:229–240.
  29. McGill W.J. (1954) Multivariate information transmission. Psychometrica 19:97–116.[Medline]
  30. Moore J.H., Gilbert J.C., Tsai C.T., Chiang F.T., Holden T., Barney N., White B.C. (2006) A flexible computational framework for detecting, characterizing, and interpreting statistical patterns of epistasis in genetic studies of human disease susceptibility. J. Theor. Biol. 241:252–261.[CrossRef][ISI][Medline]
  31. Demsar J. and Zupan B. (2004) Orange: from experimental machine learning to interactive data mining. Omnipress.
  32. Hung R.J., Hall J., Brennan P., Boffetta P. (2005) Genetic polymorphisms in the base excision repair pathway and cancer risk: a HuGE review. Am. J. Epidemiol. 162:925–942.[Abstract/Free Full Text]
  33. Manuguerra M., Saletta F., Karagas M.R., Berwick M., Veglia F., Vineis P., Matullo G. (2006) XRCC3 and XPD/ERCC2 single nucleotide polymorphisms and the risk of cancer: a HuGE Review. Am. J. Epidemiol. 164:297–302.[Abstract/Free Full Text]
  34. Hadi M.Z., Coleman M.A., Fidelis K., Mohrenweiser H.W., Wilson D.M. 3rd. (2000) Functional characterization of Ape1 variants identified in the human population. Nucleic Acids Res. 28:3871–3879.[Abstract/Free Full Text]
  35. Au W.W., Salama S.A., Sierra-Torres C.H. (2003) Functional characterization of polymorphisms in DNA repair genes using cytogenetic challenge assays. Environ. Health Perspect. 111:1843–1850.[ISI][Medline]
  36. Lotta T., Vidgren J., Tilgmann C., Ulmanen I., Melen K., Julkunen I., Taskinen J. (1995) Kinetics of human soluble and membrane-bound catechol O-methyltransferase: a revised mechanism and description of the thermolabile variant of the enzyme. Biochemistry 34:4202–4210.[CrossRef][Medline]
  37. Martone T., Vineis P., Malaveille C., Terracini B. (2000) Impact of polymorphisms in xeno(endo)biotic metabolism on pattern and frequency of p53 mutations in bladder cancer. Mutat Res. 462:303–309.[CrossRef][ISI][Medline]
  38. Hung R.J., Boffetta P., Brennan P., Malaveille C., Gelatti U., Placidi D., Carta A., Hautefeuille A., Porru S. (2004) Genetic polymorphisms of MPO, COMT, MnSOD, NQO1, interactions with environmental exposures and bladder cancer risk. Carcinogenesis 25:973–978.[Abstract/Free Full Text]
  39. Iwaki H., Kageyama S., Isono T., Wakabayashi Y., Okada Y., Yoshimura K., Terai A., Arai Y., Iwamura H., Kawakita M., et al. (2004) Diagnostic potential in bladder cancer of a panel of tumor markers (calreticulin, gamma-synuclein, and catechol-o-methyltransferase) identified by proteomic analysis. Cancer Sci. 95:955–961.[CrossRef][Medline]
  40. Yamada K., Chen Z., Rozen R., Matthews R.G. (2001) Effects of common polymorphisms on the properties of recombinant human methylenetetrahydrofolate reductase. Proc. Natl Acad. Sci. USA 98:14853–14858.[Abstract/Free Full Text]
  41. Kimura F., Florl A.R., Steinhoff C., Golka K., Willers R., Seifert H.H., Schulz W.A. (2001) Polymorphic methyl group metabolism genes in patients with transitional cell carcinoma of the urinary bladder. Mutat Res. 458:49–54.[Medline]
  42. Karagas M.R., Park S., Nelson H.H., Andrew A.S., Mott L., Schned A., Kelsey K.T. (2005) Methylenetetrahydrofolate reductase (MTHFR) variants and bladder cancer: a population-based case-control study. Int. J. Hyg. Environ. Health 208:321–327.[CrossRef][ISI][Medline]
  43. Sanyal S., Festa F., Sakano S., Zhang Z., Steineck G., Norming U., Wijkstrom H., Larsson P., Kumar R., Hemminki K. (2004) Polymorphisms in DNA repair and metabolic genes in bladder cancer. Carcinogenesis 25:729–734.[Abstract/Free Full Text]
  44. Lin J., Spitz M.R., Wang Y., Schabath M.B., Gorlov I.P., Hernandez L.M., Pillow P.C., Grossman H.B., Wu X. (2004) Polymorphisms of folate metabolic genes and susceptibility to bladder cancer: a case-control study. Carcinogenesis 25:1639–1647.[Abstract/Free Full Text]
  45. Peluso M., Airoldi L., Magagnotti C., Fiorini L., Munnia A., Hautefeuille A., Malaveille C., Vineis P. (2000) White blood cell DNA adducts and fruit and vegetable consumption in bladder cancer. Carcinogenesis 21:183–187.[Abstract/Free Full Text]
  46. Garcia-Closas M., Malats N., Silverman D., Dosemeci M., Kogevinas M., Hein D.W., Tardon A., Serra C., Carrato A., Garcia-Closas R. (2005) NAT2 slow acetylation, GSTM1 null genotype, and risk of bladder cancer: results from the Spanish Bladder Cancer Study and meta-analyses. Lancet 366:649–659.[CrossRef][ISI][Medline]
  47. Kiyohara C., Yoshimasu K., Takayama K., Nakanishi Y. (2005) NQO1, MPO, and the risk of lung cancer: a HuGE review. Genet. Med. 7:463–478.[ISI][Medline]
  48. Wu G., Jiang X., Lee W.H., Chen P.L. (2003) Assembly of functional ALT-associated promyelocytic leukemia bodies requires Nijmegen Breakage Syndrome 1. Cancer Res. 63:2589–2595.[Abstract/Free Full Text]
  49. Tennyson R.B., Ebran N., Herrera A.E., Lindsley J.E. (2002) A novel selection system for chromosome translocations in Saccharomyces cerevisiae. Genetics 160:1363–1373.[Abstract/Free Full Text]
  50. Salomon-Nguyen F., Brizard F., Le Coniat M., Radford I., Berger R., Brizard A. (1998) Abnormalities of the short arm of chromosome 12 in T cell prolymphocytic leukemia. Leukemia 12:972–975.[CrossRef][ISI][Medline]
  51. Gao N., Rahmani M., Shi X., Dent P., Grant S. (2006) Synergistic antileukemic interactions between 2-medroxyestradiol (2-ME) and histone deacetylase inhibitors involve Akt down-regulation and oxidative stress. Blood 107:241–249.[Abstract/Free Full Text]
  52. Okada N., Hirata A., Murakami Y., Shoji M., Sakagami H., Fujisawa S. (2005) Induction of cytotoxicity and apoptosis and inhibition of cyclooxygenase-2 gene expression by eugenol-related compounds. Anticancer Res. 25:3263–3269.[ISI][Medline]
  53. Clavel J., Bellec S., Rebouissou S., Menegaux F., Feunteun J., Bonaiti-Pellie C., Baruchel A., Kebaili K., Lambilliotte A., Leverger G., et al. (2005) Childhood leukaemia, polymorphisms of metabolism enzyme genes, and interactions with maternal tobacco, coffee and alcohol consumption during pregnancy. Eur. J. Cancer Prev. 14:531–540.[CrossRef][ISI][Medline]
  54. Gallegos-Arreola M.P., Batista-Gonzalez C.M., Delgado-Lamas J.L., Figuera L.E., Puebla-Perez A.M., Arnaud-Lopez L., Peralta-Leal V., Ramirez-Jirano L.J., Zuniga-Gonzalez G.M. (2004) Cytochrome P4501A1 polymorphism is associated with susceptibility to acute lymphoblastic leukemia in adult Mexican patients. Blood Cells Mol. Dis. 33:326–329.[CrossRef][ISI][Medline]
  55. Joseph T., Kusumakumary P., Chacko P., Abraham A., Radhakrishna Pillai M. (2004) Genetic polymorphism of CYP1A1, CYP2D6, GSTM1 and GSTT1 and susceptibility to acute lymphoblastic leukaemia in Indian children. Pediatr. Blood Cancer 43:560–567.[CrossRef][ISI][Medline]
  56. Krajinovic M., Richer C., Sinnett H., Labuda D., Sinnett D. (2000) Genetic polymorphisms of N-acetyltransferases 1 and 2 and gene-gene interaction in the susceptibility to childhood acute lymphoblastic leukemia. Cancer Epidemiol. Biomarkers Prev. 9:557–562.[Abstract/Free Full Text]
Received May 9, 2006; revised August 25, 2006; accepted August 25, 2006.


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 Integr Cancer TherHome page
Jianhua Wang, Yuesheng Xia, Huan Wang, and Zengxia Hou
Chinese Herbs of Shenghe Powder Reverse Multidrug Resistance of Gastric Carcinoma SGC-7901
Integr Cancer Ther, December 1, 2007; 6(4): 400 - 404.
[Abstract] [PDF]

Home page
 BioinformaticsHome page
J. K. Estrada-Gil, J. C. Fernandez-Lopez, E. Hernandez-Lemus, I. Silva-Zolezzi, A. Hidalgo-Miranda, G. Jimenez-Sanchez, and E. E. Vallejo-Clemente
GPDTI: A Genetic Programming Decision Tree Induction method to find epistatic effects in common complex diseases
Bioinformatics, July 1, 2007; 23(13): i167 - i174.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Cancer Res.Home page
D. F. Giachino, P. Ghio, S. Regazzoni, G. Mandrile, S. Novello, G. Selvaggi, D. Gregori, M. DeMarchi, and G. V. Scagliotti
Prospective Assessment of XPD Lys751Gln and XRCC1 Arg399Gln Single Nucleotide Polymorphisms in Lung Cancer
Clin. Cancer Res., May 15, 2007; 13(10): 2876 - 2881.
[Abstract] [Full Text] [PDF]

Home page
 Am. J. Respir. Crit. Care Med.Home page
T. S. Nawrot, A. Nemmar, and B. Nemery
Update in Environmental and Occupational Medicine 2006
Am. J. Respir. Crit. Care Med., April 15, 2007; 175(8): 758 - 762.
[Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
28/2/414    most recent
bgl159v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (8)
Right arrowRequest Permissions
Google Scholar
Right arrow