Associative memory by collective regulation of non-coding RNA

TL;DR

This paper proposes a biochemical network model using non-coding RNA that functions as an associative memory, capable of storing and retrieving multiple patterns, with robustness to mutations, challenging traditional views on evolutionary constraint.

Contribution

It introduces a novel biochemical network architecture that mimics neural associative memory, demonstrating its ability to store and retrieve multiple patterns through ncRNA interactions.

Findings

The network's steady state mimics associative memory neural models.

Multiple patterns can be stored simultaneously in the ncRNA concentration.

The model converges to original patterns even from different initial states.

Abstract

The majority of mammalian genomic transcripts do not directly code for proteins and it is currently believed that most of these are not under evolutionary constraint. However given the abundance non-coding RNA (ncRNA) and its strong affinity for inter-RNA binding, these molecules have the potential to regulate proteins in a highly distributed way, similar to artificial neural networks. We explore this analogy by devising a simple architecture for a biochemical network that can function as an associative memory. We show that the steady state solution for this chemical network has the same structure as an associative memory neural network model. By allowing the choice of equilibrium constants between different ncRNA species, the concentration of unbound ncRNA can be made to follow any pattern and many patterns can be stored simultaneously. The model is studied numerically and within certain parameter regimes it functions as predicted. Even if the starting concentration pattern is quite different, it is shown to converge to the original pattern most of the time. The network is also robust to mutations in equilibrium constants. This calls into question the criteria for deciding if a sequence is under evolutionary constraint.