Random Hypergraph Models of Learning and Memory in Biomolecular Networks: Shorter-Term Adaptability vs. Longer-Term Persistency

Recent progress in genomics and proteomics makes it possible to understand the biological networks at the systems level. We aim to develop computational models of learning and memory inspired by the biomolecular networks embedded in their environment. One fundamental question is how the systems rapidly adapt to their changing environment in a short period (learning) while performing persistently through the longer time span (memory). We study this issue in a probabilistic hypergraph model called the hypernetworks. The hypernetwork architecture consists of a huge number of randomly sampled hyperedges, each corresponding to higher-order micromodules in the input. We find that a system consisting of a large number of a wide range of heterogeneous low-dimensional components has a fairly competitive chance of long-term survival (memory, persistency) and short-term performance (learning, adaptability) as opposed to a system consisting of a small number of high-dimensional, fine-tuned, complex components. Empirical evidence is offered to support these findings and theoretical explanations are given