Computational mechanisms in genetic regulation by RNA

TL;DR

This paper proposes a model where non-coding RNA molecules perform neural network-like computations through their binding dynamics, suggesting a new mechanism for complex genetic regulation with high mutation tolerance.

Contribution

It introduces a simple, RNA-based computational mechanism that mimics neural networks, linking molecular interactions to information processing in genetic regulation.

Findings

RNA interactions can implement neural network algorithms.

Equilibrium constants encode stored patterns and input-output relations.

The mechanism is robust to mutations, allowing high mutation rates.

Abstract

The evolution of the genome has led to very sophisticated and complex regulation. Because of the abundance of non-coding RNA (ncRNA) in the cell, different species will promiscuously associate with each other, suggesting collective dynamics similar to artificial neural networks. Here we present a simple mechanism allowing ncRNA to perform computations equivalent to neural network algorithms such as Boltzmann machines and the Hopfield model. The quantities analogous to the neural couplings are the equilibrium constants between different RNA species. The relatively rapid equilibration of RNA binding and unbinding is regulated by a slower process that degrades and creates new RNA. The model requires that the creation rate for each species be an increasing function of the ratio of total to unbound RNA. Similar mechanisms have already been found to exist experimentally for ncRNA regulation. With the overall concentration of RNA regulated, equilibrium constants can be chosen to store many different patterns, or many different input-output relations. The network is also quite insensitive to random mutations in equilibrium constants. Therefore one expects that this kind of mechanism will have a much higher mutation rate than ones typically regarded as being under evolutionary constraint.