Mitsuo Yoshida

On activation functions for complex-valued neural networks: existence of energy functions

Recently models of neural networks that can directly deal with complex numbers, complex-valued neural networks, have been proposed and several studies on their abilities of information processing have been done. One of the important factors to characterize behavior of a complex-valued neural network is its activation function which is a nonlinear complex function. This paper discusses the properties of activation functions from the standpoint of existence of an energy function for complex-valued neural networks. Two classes of complex functions which are widely used as activation functions in the models of complex-valued neural networks are considered. We investigate the properties of activation functions which assure existence of energy functions and discuss about how to find out complex functions which satisfy the properties.

Models of Hopfield-type quaternion neural networks and their energy functions.

Recently models of neural networks that can directly deal with complex numbers, complex-valued neural networks, have been proposed and several studies on their abilities of information processing have been done. Furthermore models of neural networks that can deal with quaternion numbers, which is the extension of complex numbers, have also been proposed. However they are all multilayer quaternion neural networks. This paper proposes models of fully connected recurrent quaternion neural networks, Hopfield-type quaternion neural networks. Since quaternion numbers are non-commutative on multiplication, some different models can be considered. We investigate dynamics of these proposed models from the point of view of the existence of an energy function and derive their conditions for existence.

Enzymatic Repair of 5-Formyluracil II. MISMATCH FORMATION BETWEEN 5-FORMYLURACIL AND GUANINE DURING DNA REPLICATION AND ITS RECOGNITION BY TWO PROTEINS INVOLVED IN BASE EXCISION REPAIR (AlkA) AND MISMATCH REPAIR (MutS)

5-Formyluracil (fU), a major methyl oxidation product of thymine, forms correct (fU:A) and incorrect (fU:G) base pairs during DNA replication. In the accompanying paper (Masaoka, A., Terato, H., Kobayashi, M., Honsho, A., Ohyama, Y., and Ide, H. (1999)J. Biol. Chem. 274, 25136–25143), it has been shown that fU correctly paired with A is recognized by AlkA protein (Escherichia coli 3-methyladenine DNA glycosylase II). In the present work, mispairing frequency of fU with G and cellular repair protein that specifically recognized fU:G mispairs were studied using defined oligonucleotide substrates. Mispairing frequency of fU was determined by incorporation of 2′-deoxyribonucleoside 5′-triphosphate of fU opposite template G using DNA polymerase I Klenow fragment deficient in 3′-5′ exonuclease. Mispairing frequency of fU was dependent on the nearest neighbor base pair in the primer terminus and 2–12 times higher than that of thymine at pH 7.8 and 2.6–6.7 times higher at pH 9.0 with an exception of the nearest neighbor T(template):A(primer). AlkA catalyzed the excision of fU placed opposite G, as well as A, and the excision efficiencies of fU for fU:G and fU:A pairs were comparable. In addition, MutS protein involved in methyl-directed mismatch repair also recognized fU:G mispairs and bound them with an efficiency comparable to T:G mispairs, but it did not recognize fU:A pairs. Prior complex formation between MutS and a heteroduplex containing an fU:G mispair inhibited the activity of AlkA to fU. These results suggest that fU present in DNA can be restored by two independent repair pathways, i.e. the base excision repair pathway initiated by AlkA and the methyl-directed mismatch repair pathway initiated by MutS. Biological relevance of the present results is discussed in light of DNA replication and repair in cells.

A Model of Hopfield-Type Quaternion Neural Networks and Its Energy Function

Recently models of neural networks that can directly deal with complex numbers, complex-valued neural networks, have been proposed and several studies on their abilities of information processing have been done. Furthermore models of neural networks that can deal with quaternion numbers, which is an extension of complex numbers, have also been proposed. However they are all multilayer quaternion neural networks. This paper proposes a model of recurrent quaternion neural networks, Hopfield-type quaternion neural networks. We investigate dynamics of the proposed model from the point of view of the existence of an energy function and derive its condition.

The Effect of Tetrahydroisoquinoline on the Mitochondrial Respiration

1-Methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP)-induced parkinsonism has been considered as the best model available at present of Parkinson’s disease (1–4). The neurotoxic effect of MPTP on the nigrostriatal dopaminergic neurons resides in the inhibition of the mitochondrial respiration by 1-methyl-4-phenylpyridinium ion (MPP+), an oxidation product of MPTP, resulting in the energy crisis of those neurons (5–11). Since the discovery of MPTP and MPP+, it has been proposed that (an) exogenous or (an) endogenous MPTP-like substance(s) may be the cause of Parkinson’s disease. 1,2,3,4-Tetrahydroisoquinoline (TIQ) has emerged as one of such candidates (12). TIQ was shown to inhibit the state 3 respiration of the mitochondria (13). This effect of TIQ on the mitochondrial respiration is similar to that of MPP+. In this communication, we report effects of TIQ on the enzyme-protein complexes in the electron transport system and on the respiratory enzymes in the tricarboxylic acid (TCA) cycle.