Using Glauber-Pseudolikelihood Dynamics to Generate Artificial Proteins

Since the fundamental processes governing the functioning of the biological cell were first discovered, it became clearer and clearer as research on the subject continued in time that proteins are responsible for a numerous ensemble of different activities which are essential to the survival of all life forms present on this planet. Given this fact, it goes without saying that one of the first issues that biologists and scientists had to deal with was to analyze proteins in order to identify their functions inside the organism. It is well known that proteins are macro-molecules constituted by organic monomers, the amino acids, linked in a chain that eventually, after its assembly in the cell, proceeds to fold into a particular three-dimensional structure, known as the tertiary structure of proteins. One of the most remarkable discoveries on the subject is that it is indeed this configuration, conveyed by inter-molecular forces between amino acids, that is responsible for the specific role of proteins inside the organism. After examining several proteins by the means of x-ray spectroscopy, discerning their amino acidic sequence and structure and detecting their functionality, it was learned by the scientific community that many proteins obtained from even very different types of organism share a similar three-dimensional pattern. The most endorsed hypothesis is that these proteins have a common evolutionary origin, and while the mutation mechanisms at the basis of evolution gave rise in the millions of years since the emergence of life on this planet to the observable amino acids' variability seen in these sequences, the natural selection instead preserved their structure almost unaltered. It was then straightforward to collect all these proteins, called homologous, inside different families according to their function. While the problem of classifying proteins inside evolutionary-related families has been successfully solved since the 1970', in more recent years a more demanding challenge has arised: that is to biosynthesize artificial sequences capable of mimicking a desired function associated with a certain protein family, and at the same time being more stable and efficient than the natural ones. On the basis of the works of E.Aurell and S.Cocco this paper aims at showing that it is possible to design a generative algorithm capable of building up original amino acids' sequences using parameters inferred from various protein domain families, with the purpose of reproducing the distinguishing three-dimensional structure and therefore the function of the family analyzed if eventually assembled in laboratory. An illustration of the way it is possible to infer these parameters from the protein families and how they serve the purpose of generating these sequences will follow in the upcoming sections of this work.