Abstract
DNAzymes are short pieces of DNA with catalytic activity, capable of cleaving RNA. DNAzymes have multiple applications as biosensors and in therapeutics. The high specificity and low toxicity of these molecules make them particularly suitable as therapeutics, and clinical trials have shown that they are effective in patients. However, the development of DNAzymes has been limited due to the lack of specific tools to identify efficient molecules, and users often resort to time-consuming/costly large-scale screens. Here, we propose a computational methodology to identify 10-23 DNAzymes that can be used to triage thousands of potential molecules, specific to a target RNA, to identify those that are predicted to be efficient. The method is based on a logistic regression and can be trained to incorporate additional DNAzyme efficiency data, improving its performance with time. We first trained the method with published data, and then we validated, and further refined it, by testing additional newly synthesized DNAzymes in the laboratory. We found that although binding free energy between the DNAzyme and its RNA target is the primary determinant of efficiency, other factors such as internal structure of the DNAzyme also have an important effect. A program implementing the proposed method is publicly available.