Artificial neural networks methods for identification of the most relevant genes from gene expression array data

Boger, Zvi

doi:10.1109/ijcnn.2003.1224066

Cited by 13 publications

(7 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The first argument about ANNs is that the modeled networks are ove he weights learned are hard to be understood by humans compared with other rule-based classifiers [63,84]. Although it is computationally difficult, the knowledge about the weights learned could be retrieved in the form of Causal Indices (CI) [84]. Another shortcoming about ANNs is that they will over-fit the training data and generalize to new data poorly when the sample size is small.…”

Section: Discussionmentioning

confidence: 99%

Computational Hybrid Systems for Identifying Prognostic Gene Markers of Lung Cancer

Wan¹

View full text Add to dashboard Cite

show abstract

Section: Discussionmentioning

confidence: 99%

Computational Hybrid Systems for Identifying Prognostic Gene Markers of Lung Cancer

Wan¹

View full text Add to dashboard Cite

show abstract

“…Boger [44] demonstrated the application of artificial neural networks for gene array analysis and cancer cell identification. The data was first trained using a Principal Component Analysis training algorithm and then local minima avoidance and escape algorithms were used.…”

Section: Artificial Neural Networkmentioning

confidence: 99%

“…Neural networks are difficult to understand and are not easily extensible. Neural networks are considered to be black boxes [44] as the process that is going on in the hidden layer is not known to the user. Moreover many applications of artificial neural networks include classification analyses but to our knowledge, there are no applications for the complete modeling of gene-gene interactions yet.…”

Section: Artificial Neural Networkmentioning

confidence: 99%

Identifying prognostic gene-signatures using a network-based approach

Nutakki¹

View full text Add to dashboard Cite

Identifying Prognostic Gene Signatures Using a Network-Based Approach Swetha Bose Nutakki The main objective of this study is to develop a novel network-based methodology to identify prognostic signatures of genes that can predict recurrence in cancer. Feature selection algorithms were used widely for the identification of gene signatures in genome-wide association studies. But most of them do not discover the causal relationships between the features and need to compromise between accuracy and complexity. The network-based techniques take the molecular interactions between pairs of genes in to account and are thus a more efficient means of finding gene signatures, and they are also better in terms of its classification accuracy without compromising over complexity. Nevertheless, the network-based techniques currently being used have a few limitations each. Correlation-based coexpression networks do not provide predictive structure or causal relations among the genes. Bayesian networks cannot model feedback loops. Boolean networks can model small scale molecular networks, but not at the genome-scale. Thus the prediction logic induced implication networks are chosen to generate genome-wide coexpression networks, as they integrate formal logic and statistics and also overcome the limitations of other network-based techniques. The first part of the study includes building of an implication network and identification of a set of genes that could form a prognostic signature. The data used consisted of 442 samples taken from 4 different sources. The data was split into training set UM/HLM (n=256) and two testing sets DFCI (n=82) and MSK (n=104). The training set was used for the generation of the implication network and eventually the identification of the prognostic signature. The test sets were used for validating the obtained signature. The implication networks were built by using the gene expression data associated with two disease states (metastasis or non-metastasis), defined by the period and status of post-operative survival. The gene interactions that differentiated the two disease states, the differential components, were identified. The major cancer hallmarks (E2F, EGF, EGFR, KRAS, MET, RB1, and TP53) were considered, and the genes that interacted with all the major hallmarks were identified from the differential components to form a 31-gene prognostic signature. A software package was created in R to automate this process which has C-code embedded into it. Next, the signature was fitted into a COX proportional hazard model and the nearest point to the perfect classification in the ROC curve was identified as the best scheme for patient stratification on the training set (log-rank p-value =1.97e-08), and two test sets DFCI (log-rank pvalue =2.13e-05) and MSK (log-rank p-value = 1.24e-04) in Kaplan-Meier analyses. Prognostic validation was carried out on the test sets using methods such as Concordance Probability Estimate (CPE) and Gene Set Enrichment Analysis (GSEA). The accuracy of this signature was evalua...

show abstract

“…1. The optimization of a function di erent from the covariance, usually the sumof-squares error Littmann and Ritter 1992a,b;Lehtokangas 1999Lehtokangas , 2000 Ma and Khorasani 2000, 2003, 2004 . New hidden layers also can be added in Ma and Khorasani 2003 .…”

Section: Cascade Correlationmentioning

confidence: 99%

Learning with Feed-forward Neural Networks: Three Schemes to Deal with the Bias/Variance Trade-off

Romero Merino

View full text Add to dashboard Cite

In terms of the Bias/Variance decomposition, very flexible (i.e., complex) Supervised Machine Learning systems may lead to unbiased estimators but with high variance. A rigid model, in contrast, may lead to small variance but high bias. There is a trade-off between the bias and variance contributions to the error, where the optimal performance is achieved. In this work we present three schemes related to the control of the Bias/Variance decomposition for Feed-forward Neural Networks (FNNs) with the (sometimes modified) quadratic loss function: 1. An algorithm for sequential approximation with FNNs, named Sequential Approximation with Optimal Coefficients and Interacting Frequencies (SAOCIF). Most of the sequential approximations proposed in the literature select the new frequencies (the non-linear weights) guided by the approximation of the residue of the partial approximation. We propose a sequential algorithm where the new frequency is selected taking into account its interactions with the previously selected ones. The interactions are discovered by means of their optimal coefficients (the linear weights). A number of heuristics can be used to select the new frequencies. The aim is that the same level of approximation may be achieved with less hidden units than if we only try to match the residue as best as possible. In terms of the Bias/Variance decomposition, it will be possible to obtain simpler models with the same bias. The idea behind SAOCIF can be extended to approximation in Hilbert spaces, maintaining orthogonal-like properties. In this case, the importance of the interacting frequencies lies in the expectation of increasing the rate of approximation. Experimental results show that the idea of interacting frequencies allows to construct better approximations than matching the residue. 2. A study and comparison of different criteria to perform Feature Selection (FS) with Multi-Layer Perceptrons (MLPs) and the Sequential Backward Selection (SBS) procedure within the wrapper approach. FS procedures control the Bias/Variance decomposition by means of the input dimension, establishing a clear connection with the curse of dimensionality. Several critical decision points are studied and compared. First, the stopping criterion. Second, the data set where the value of the loss function is measured. Finally, we also compare two ways of computing the saliency (i.e., the relative importance) of a feature: either first train a network and then remove temporarily every feature or train a different network with every feature temporarily removed. The experiments are performed for linear and non-linear models. Experimental results suggest that the increase in the computational cost associated with retraining a different network with every feature temporarily removed previous to computing the saliency can be rewarded with a significant performance improvement, specially if non-linear models are used. Although this idea could be thought as very intuitive, it has been hardly used in practice. Regarding the data set where the value of the loss function is measured, it seems clear that the SBS procedure for MLPs takes profit from measuring the loss function in a validation set. A somewhat non-intuitive conclusion is drawn looking at the stopping criterion, where it can be seen that forcing overtraining may be as useful as early stopping. 3. A modification of the quadratic loss function for classification problems, inspired in Support Vector Machines (SVMs) and the AdaBoost algorithm, named Weighted Quadratic Loss (WQL) function. The modification consists in weighting the contribution of every example to the total error. In the linearly separable case, the solution of the hard margin SVM also minimizes the proposed loss function. The hardness of the resulting solution can be controlled, as in SVMs, so that this scheme may also be used for the non-linearly separable case. The error weighting proposed in WQL forces the training procedure to pay more attention to the points with a smaller margin. Therefore, variance tries to be controlled by not attempting to overfit the points that are already well classified. The model shares several properties with the SVMs framework, with some additional advantages. On the one hand, the final solution is neither restricted to have an architecture with so many hidden units as points (or support vectors) in the data set nor to use kernel functions. The frequencies are not restricted to be a subset of the data set. On the other hand, it allows to deal with multiclass and multilabel problems in a natural way. Experimental results are shown confirming these claims. A wide experimental work has been done with the proposed schemes, including artificial data sets, well-known benchmark data sets and two real-world problems from the Natural Language Processing domain. In addition to widely used activation functions, such as the hyperbolic tangent or the Gaussian function, other activation functions have been tested. In particular, sinusoidal MLPs showed a very good behavior. The experimental results can be considered as very satisfactory. The schemes presented in this work have been found to be very competitive when compared to other existing schemes described in the literature. In addition, they can be combined among them, since they deal with complementary aspects of the whole learning process.

show abstract

Artificial neural networks methods for identification of the most relevant genes from gene expression array data

Cited by 13 publications

References 21 publications

Computational Hybrid Systems for Identifying Prognostic Gene Markers of Lung Cancer

Computational Hybrid Systems for Identifying Prognostic Gene Markers of Lung Cancer

Identifying prognostic gene-signatures using a network-based approach

Learning with Feed-forward Neural Networks: Three Schemes to Deal with the Bias/Variance Trade-off

Contact Info

Product

Resources

About