Recursive Neural Network Model Research Articles

Forecasting gold price changes: Rolling and recursive neural network models

This paper analyzes recursive and rolling neural network models to forecast one-step-ahead sign variations in gold price. Different combinations of techniques and sample sizes are studied for feed forward and ward neural networks. The results shows the rolling ward networks exceed the recursive ward networks and feed forward networks in forecasting gold price sign variation. The results support the use of neural networks with a dynamic framework to forecast the gold price sign variations, recalculating the weights of the network on a period-by-period basis, through a rolling process. Our results are validated using the block bootstrap methodology with an average sign prediction of 60.68% with a standard deviation of 2.82% for the rolling ward net.

Ionic liquids: prediction of their melting points by a recursive neural network model

A recursive neural network (RNN) was used to predict the melting points of several pyridinium-based ionic liquids (ILs). The RNN is a neural network model for processing structured data that allows for the direct handling of chemical compounds as labelled rooted ordered trees. It constitutes a direct approach to quantitative structure–property relationship (QSPR) of ILs, which avoids the use of dedicated molecular descriptors. The adopted representation of molecular structures captures significant topological aspects and chemical functionalities for each molecule in a general and flexible way. Particular emphasis was given to the representation of cyclic moieties. The model was applied to a set of 126 pyridinium bromides; it was validated by splitting the dataset into a disjoint training set (100 compounds) and test set (26 compounds). Comparison with the results obtained by other QSPR approaches on the same dataset is also presented.

Predicting Physical−Chemical Properties of Compounds from Molecular Structures by Recursive Neural Networks

In this paper, we report on the potential of a recently developed neural network for structures applied to the prediction of physical chemical properties of compounds. The proposed recursive neural network (RecNN) model is able to directly take as input a structured representation of the molecule and to model a direct and adaptive relationship between the molecular structure and target property. Therefore, it combines in a learning system the flexibility and general advantages of a neural network model with the representational power of a structured domain. As a result, a completely new approach to quantitative structure-activity relationship/quantitative structure-property relationship (QSPR/QSAR) analysis is obtained. An original representation of the molecular structures has been developed accounting for both the occurrence of specific atoms/groups and the topological relationships among them. Gibbs free energy of solvation in water, Delta(solv)G degrees , has been chosen as a benchmark for the model. The different approaches proposed in the literature for the prediction of this property have been reconsidered from a general perspective. The advantages of RecNN as a suitable tool for the automatization of fundamental parts of the QSPR/QSAR analysis have been highlighted. The RecNN model has been applied to the analysis of the Delta(solv)G degrees in water of 138 monofunctional acyclic organic compounds and tested on an external data set of 33 compounds. As a result of the statistical analysis, we obtained, for the predictive accuracy estimated on the test set, correlation coefficient R = 0.9985, standard deviation S = 0.68 kJ mol(-1), and mean absolute error MAE = 0.46 kJ mol(-1). The inherent ability of RecNN to abstract chemical knowledge through the adaptive learning process has been investigated by principal components analysis of the internal representations computed by the network. It has been found that the model recognizes the chemical compounds on the basis of a nontrivial combination of their chemical structure and target property.

Genomics, morphogenesis and biophysics: Triangulation of Purkinje cell development

The cerebellar Purkinje cells (P-cells) comprise an organelle that is suitable for combined analysis by morphology and genomics, using biophysical tools. In some unknown way, genomic information specifies the development of P-cells. One of us (AJP) has previously proposed that fractal processes associated with DNA are in a causal relation to the fractal properties of organelles such as P-cells (FractoGene, 2002, patent pending). This fractal postulate predicts that the dendritic arborization of P-cells will be less complex in lower order vertebrates. The prediction can be tested by systematic comparative neuroanatomy of the P-cell in species for which genome sequences permit inter-species comparison. The Fugu rubripes (Fugu), Danio rerio (Danio) and other species are lower order vertebrates for which genome sequences are available and tests could be conducted. Consistent with the fractal prediction, P-cell dendritic arbor is primitive in Fugu, being much less complex than in Mus musculus and in Homo sapiens. Genomic analysis readily identified PEP19/Pcp4, Calbindin-D28k, and GAD67 genes in Fugu and in Danio that are closely associated with P-cells in Canis familiaris, Rattus norvegicus, Mus musculus and Homo sapiens. Gene L7/Pcp2 exhibits strongest association with P-cells in higher vertebrates. L7/Pcp2 shows strong protein residue homology with genes greater than 600 residues and including 2-3 GoLoco domains, designated as having G protein signaling modulator function (AGS3-like proteins). Fugu has a short gene with a single GoLoco domain, but it has greatest homology with the AGS3-like proteins. No similar short gene is present in Danio or in Xenopus. Classical L7/Pcp2 is only detected in higher vertebrates, suggesting that it may be a marker of more recent evolutionary development of cerebellar P-cells. We expect that a new generation of data mining tools will be required to support recursive fractal geometrical, combinatorial, and neural network models of the genomic basis of morphogenesis.

Recursive neural networks for processing graphs with labelled edges: theory and applications

In this paper, we introduce a new recursive neural network model able to process directed acyclic graphs with labelled edges. The model uses a state transition function which considers the edge labels and is independent both from the number and the order of the children of each node. The computational capabilities of the new recursive architecture are assessed. Moreover, in order to test the proposed architecture on a practical challenging application, the problem of object detection in images is also addressed. In fact, the localization of target objects is a preliminary step in any recognition system. The proposed technique is general and can be applied in different detection systems, since it does not exploit any a priori knowledge on the particular problem. Some experiments on face detection, carried out on scenes acquired by an indoor camera, are reported, showing very promising results.

Prediction of cystine connectivity using SVM.

One of the major contributors to protein structures is the formation of disulphide bonds between selected pairs of cysteines at oxidized state. Prediction of such disulphide bridges from sequence is challenging given that the possible combination of cysteine pairs as the number of cysteines increases in a protein. Here, we describe a SVM (support vector machine) model for the prediction of cystine connectivity in a protein sequence with and without a priori knowledge on their bonding state. We make use of a new encoding scheme based on physico-chemical properties and statistical features (probability of occurrence of each amino acid residue in different secondary structure states along with PSI-blast profiles). We evaluate our method in SPX (an extended dataset of SP39 (swiss-prot 39) and SP41 (swiss-prot 41) with known disulphide information from PDB) dataset and compare our results with the recursive neural network model described for the same dataset.

Modified Herault–Jutten Algorithms for Blind Separation of Sources

We present several modifications of blind separation adaptive algorithms which have significant advantages over the well-known Herault–Jutten learning algorithm in handling ill-conditioned signals. In particular, the proposed algorithms are more stable and converge to the correct solutions in cases where previous algorithms did not. The problem is the classical one in which several independent source signals s j ( t) ( j = 1, 2,…, n) are linearly combined via unknown mixing coefficients (parameters) a ij to form observations x i ( t) = ∑ j=1 n a ij s j ( t), i= 1, 2, …, n. The synaptic weights w ij of a linear system (often referred to as a single-layer feedforward neural network) must be adapted to combine the observations x i ( t) to form optimal estimates of the source signals ŝ p ( t) = y p ( t) = ∑ i=1 n w pix i ( t). The optimal weights correspond to the statistical independence of the output signals y p ( t) and they simultaneously ensure self-normalization of these signals. Starting from the modified Herault–Jutten recursive neural network model, we have derived a family of on-line adaptive learning algorithms for feedback (fully recurrent) and feedforward architectures. The validity and high performance of the proposed neural network are illustrated by simulation.