Computing with Nucleic Acids

Exactly one decade has passed since Adleman’s seminal demonstration that biochemical operations can be performed on DNA in order to solve certain combinatorial problems [1]. So, the time seems ripe to critically analyze where the field stands, what has been accomplished, what needs to be accomplished, and perhaps most importantly, what are the realistic expectations in this field. A rational assessment is also pertinent because the whole field of DNA computing has come to be considered by many as being ‘‘more hype than substance’’ [2], and the practitioners in the field are partly to blame for this. Namely, initial opening in the field was so impressive and inspirational that it garnered understandable yet, in retrospect, completely unjustified exuberance about the potential applications of ‘‘DNA computers’’. As a result, it became standard in the field to put forward incredible predictions in the abstracts and conclusions of papers. This led to a situation where many very important advances in the field of DNA computing are viewed with suspicion by computer scientists and are misunderstood by biochemists. Misinterpretations in the popular press did not help either (not to mention the Schön affair in the somewhat related field of molecular electronics). In this respect, it is sobering to read the latest Science paper from Adleman [3], in which he scales the ‘‘Mount Everest’’ of DNA parallel computation (solving a 20-variable instance of 3-SAT), yet he puts the field in a realistic perspective. At the very beginning, we have to define what DNA computation is. Many examples in the literature do not compute anything tangible (in layman’s terms, at least), and in many cases, actual computation is not performed by DNA, but by a human operator. So we decided to discuss those approaches in which authors claim to be able to perform computation by using some type of DNA hybridization procedure. We will specifically exclude methods in which DNA is only incidentally used and in which, for example, Boolean behavior is actually displayed by proteins [4] or by cells. In particular, for reasons of space, we decided to focus only on those approaches that have appeared in mainstream peer-reviewed literature more than once (i.e., that established themselves), and we could not, for the most part, cover the abundant literature that appeared in various DNA Computation conferences. This review is split into four main sections according to the major thrusts in what could loosely be defined as ‘‘the field of DNA computation’’: (a) Adleman’s experiment