Improving the quality of semantic retrieval in DNA-based memories with learning

At least three types of associative memories based on DNA-affinity have been proposed. Previously, we have quantified the quality of retrieval of genomic information in simulation by comparison to state-of-the-art symbolic methods available, such as LSA (Latent Semantic Analysis.) Their ability is poor when performed without a proper compaction procedure. Here, we use a different compaction procedure that uses learning to improve the ability of DNA-based memories to store abiotic data. We evaluate and compare the quality of the retrieval of semantic information. Their performance is much closer to that of LSA, according to human expert ratings, and slightly better than the previous method using a summarization procedure. These results are expected to improve and feasibly scale up with actual DNA molecules in real test tubes. © Springer-Verlag 2004