October 1997

"Microbial Neural Network:

Artificial Intelligence from Fungi"

(A Multifunctional Biochip)

by Vladimir N. Ivanov

Faculty of Biology, Ukrainan National University

Kiev, Ukraine

ABSTRACT. Microbial chip may be created as a pseudo - neural network grown as mycelium of fungi, connected with the computer. Thic chip which is a net of branched and anastomosed hyphaes of mycelium may be grown / created under the learning electric impulses controlled by computer. Notwithstanding the low rate of fuzzy logic operations such chip may be useful for the solving the anthropomorphic problems and for the biosensoring. Theoretical and engineering issues of fungi-based chip are briefly considered in this paper.

CONCEPTUAL FRAMEWORK. The theory and simulation of neural networks (NN) are considered to be the basis for the construction of neurocomputers to solve the information problems with simultaneous influences of numerous factors and various kinds of information on the system (1-5,7). The recognition of information under uncertain and fluctuated conditions and under conditions of fuzzy reasoning is the evolutionary prerogative of the biological systems.

One way for the decision of antropomorphic problems is the use of such features of life as self replication, adaptation and evolution in the man-made electronic systems of computation.The directions of this way are the evolutionary computation and programming, cellular programming algorithms, cellular automata, biomorphological robotics, and other bio-inspired hardware and software systems (6,8-10).

Another way is the direct use of biological elements like biomolecules, cells or tissues within the electronic systems. The enzymes, bioreceptors and polynucleotides are used at present in the biosensors; the photosensitive proteins are considered as important elements of bioelectronics. The mammalian cells have been also used as the element of biosensor (3). The simplicity of cultivation of microbial cells and their great biodiversity and stability in the artificial computational systems give the reason to consider the microbial multicellular systems as the most prospective ones for the creation of biochip which can modelling the artificial intelligence. The hybridization of neural - networks - like microbiological system with a computer can give the best results for solving anthropomorphic information problems such as pattern or speech recognition.

PHYSICAL AND BIOLOGICAL BASIS - The hybrid of computer - enhanced learning and working systems with neural - networks - like mycelia of microscopic fungi may be called as mycocomputer or Fungal Artificial Intelligence (FAI). Fungi are the kingdom of heterotrophic organisms widespread in nature. Cultivation of fungal colonies in laboratory is feasible for many species of fungi by ordinary microbiological methods in liquid or solid synthetic media. The growth of filamentous fungi is the extension and branching of hyphae. A hyphae is a cylinder covered by rigid cell wall with diameter from 5 to 15 micrometers. Fungal hyphae grow in a strict polarized manner, extending only at the extreme tip, called apix. Hyphae can connect , aggregate and interlace between themselves, forming mycelium. Fungal hyphaes conduct electric currents through themselves and establish endogenous electric fields. One type of this field is the membrane poteential and other one is the lateral polarity of hyphae mainly expressed at apical region of hyphae. Electric currents into the apex and out of the trunk have been noted in a variety of fungi (4). As a consequence, the applied electric field affects the hyphae growth and branching. For the majority of fungal cells, the physiological range of steady electric fields would be within 0.1 - 10 mV per cell diameter (2). The growth of fungal colonies in the applied electric fields with intensity from 15 to 30 V/cm shows the branching and growth of hyphae towards the anade or the perpendicular growth (2). Applied electric fields may generate intracellular voltage gradients by depolarizing the membrane at the cathodic end of a cell and hyperpolarizing it at the anodic end. A cytoplasmatic electric field between the apices and distal region of hyphae is about 0.5 V/cm. The time required for polarization makes up some minutes. This time is necessary for the electrophoretical distribution of active polymers in fungal hyphae (2).

ANALOGIES BETWEEN NEURON NETWORK AND MYCELIUM. Typical neuron consists of the cell body, branching dendrites, one axon with collaterals. The neurons are connected between themselves throughout stimulating and braking synapses. The natural NN-like mycelial network also has interhyphae electric contacts. It is proposed that these contacts hyperpolarize or depolarize the membrane potential in the apical region of hyphae as the consequences of disposition of electric contacts from the apices and syncronism of polarity of electric fields applied to contacting mycelia. Individual mycelia may be considered as neuron-like biological element of neurocomputer. The difference between neuron and mycelium is in great number of mycelial exit channels. Natural neural networks and mycelial networks are very similar systems. Taking into account the technological possibility to cultivate the fungal mycelium and to connect them with any electric system, it would be interesting to create the hybrid of FAI and computer-enhanced learning and working system.

LEARNING WORK AND MEMORY OF FAI. The learning of NN is based on the increase of the connections between synchronously active neuron-like elements. The learning of FAI can be carried out by synchronization of the electric impulses from electronic or biological receptors and the impulses corresponding to right answer(s) from learning computer. Coincidence of these impulses may stimulate or inhibit the extension and branching of neighbouring hyphaes and influences on the number of interhyphal electric contacts. It can be the base of FAI memory. The growth of mycelium and memory of FAI will correlate between themselves. The duration of fungal growth and the duration of FAI learning may be about 5 days which is the time of growth cycle. After learning the growth of fungi is finished and the using of FAI may begin. The duration of FAI active work may be about 20-30 days. The mass of one mycelium is about one milligram and the information capacity of one mycelium as pseudo-neuron is about 10 bits. Thus, the specific volume of FAI memory in the biochip may be about 10,000,000 bits per 1 gram of fungal biomass. The result of FAI work is the information in the form of electric signals from the biochip which are transformed and analyzed by computer.

ENGINEERING APPROACH. The main engineering problems in the building of FAI are (i) the stability of biochip, (ii) the formation of electric contacts of sensors and computer with some thousands of inlet, outlet and inlet mycelia, (iii) the providing of fungi with oxygen and nutrients. The stabilization problem may be solved by the using of thermophilic micromycetes as the biological base. The providing of fungy with oxygen and nutrients is carried out by cultivation of fungi on porous hydrophilic carrier. The electric contacts of sensors and computers with some thousands of inlet, outlet and inlet mycelia are formed by disposition of fungal conidia in micropittings connected with electrical contacts, as well as germination of conidia throughout the liquid isolation layer between the micropittings and porous hydrophilic carrier.

REFERENCES

(1) - Encyclopedia of Artificial Intelligence, (Shapiro S.C., ed.), 2nd ed., John Wiley & Sons, New York, 1992

(2) - Gow., N.A.R. - Polarity and branching in fungi induced by electrical fields. in: Spatial organization in eukaryotic microbes (special publications of the Society for General Microbiology) 23 (1987), 25-41

(3) - Gross, G.W., Rhoades, B.K., Azzazy, H.M.E., Wu,M.C. - The use of neuronal networks on multielectrode arrays as biosensors- in: Biosensors & Bioelectronics 10 (1995), 553 - 567

(4) - Harold, F.M., Caldwell, J.H. and Schreurs, W.J.A.- Endogenous electric current and polarized bgrowth of fungal hyphae - in: Spatial organization in eukaryotic microbes (special publications of the Society for General Microbiology) 23 (1987), 11 - 23

(6) - Neural Network Systems Techniques and Applications (C.T. Leondes, ed.), Academic Press, New York, 1996

(5) Luger,G., Stubblefield, W. - Artificial Intelligence: Structures and Strategies for Complex Problem Solving - 2nd ed. The Benjamin / Cummings Publishing Company, Inc. 1993

(7) - Neurocomputers and Intellectual Robots - (Amosov, N.M., ed.) Naukova dumka , Kiev, 1991 (in Russian)

(8) - Parallel Problem Solving from Nature - (Voigt, H.M., Ebeling, W., Rechenberg, I. and SchwefelH.-P., eds.), Springer - Verlag, 1996

(9) - Pratt, I. - Artificial Intelligence - Macmillan, London, 1994

(10) - Proceedings of "The First International Conference on Evolvable Systems: from Biology to Hardware (ICES96)" - Springer-Verlag, 1997

Suggested further reading:

Watson, A. - Why can't a computer be more like a brain? - in: Science 277 (1997), 1934 - 1936

Bains, S.- A subtler silicon cell for neural networks - in: Science 277 (1997), 1935