Neurosymbolic Approach in Biological Research

Abstract

Modelling and studying the processes and methods of intercellular and intracellular signalling cascades regulation involved in the process of programmed cell death and searching for substances capable of influencing the activation or inhibition of the process of cell apoptosis and the methods of their transportation to a given cell, is one of the numerous actual and open issues in biological research. A safe and fast method for this that does not require research on living organisms is computer molecular modelling. Many approaches and tools have been proposed and developed in the last decade. In particular, today, we observe a wide use of analytical methods for drug creation and a search for effective treatment methods. Such methods include modern methods of artificial intelligence (AI) based on neural network technology and methods of modelling interactions in biological and chemical processes at different levels of abstraction. Neural networks are used to obtain the ligand representation, protein compounds, and others and to build predictive models of the molecular compound properties widely used in drug discovery research. Modelling methods for both continuous and discrete models are applied using various approaches: statistical, probabilistic, simulation, and visual. The most well-known and used molecular modelling methods include the docking method, the molecular dynamics method, and the Monte Carlo method. To date, many software tools that support these methods have been developed. However, the considered modelling approaches and tools have a number of disadvantages, which can be of critical importance for conducting experiments. This article presents a new approach to modelling biochemical processes and biological systems based on the formalism of the behaviour algebra and algebraic modelling language APLAN and its combination with neural network methods, the so-called Neurosymbolic approach. In particular, the possibility of multilevel modelling (from the level of the atomic structure of substances and quantum–mechanical interactions to the level of interaction of biological objects) and modelling of biological systems as complex hybrid systems that combine discrete and continuous processes is considered. A brief review of the current research on using neural network methods in biological research was also presented.