SequenceCraft: machine learning-based resource for exploratory analysis of RNA-cleaving deoxyribozymes

Deoxyribozymes or DNAzymes represent artificial short DNA sequences bearing many catalytic properties. In particular, DNAzymes able to cleave RNA sequences have a huge potential in gene therapy and sequence-specific analytic detection of disease markers. This activity is provided by catalytic cores able to perform site-specific hydrolysis of the phosphodiester bond of an RNA substrate. However, the vast majority of existing DNAzyme catalytic cores have low efficacy in in vivo experiments, whereas SELEX based on in vitro screening offers long and expensive selection cycle with the average success rate of ~ 30%, moreover not allowing the direct selection of chemically modified DNAzymes, which were previously shown to demonstrate higher activity in vivo. Therefore, there is a huge need in in silico approach for exploratory analysis of RNA-cleaving DNAzyme cores to drastically ease the discovery of novel catalytic cores with superior activities.

In this work, we develop a machine learning based open-source platform SequenceCraft allowing experimental scientists to perform DNAzyme exploratory analysis via quantitative observed rate constant (k_obs) estimation as well as statistical and clustering data analysis. This became possible with the development of a unique curated database of > 350 RNA-cleaving catalytic cores, property-based sequence representations allowing to work with both conventional and chemically modified nucleotides, and optimized k_obs predicting algorithm achieving Q² > 0.9 on experimental data published to date.

This work represents a significant advancement in DNAzyme research, providing a tool for more efficient discovery of RNA-cleaving DNAzymes. The SequenceCraft platform offers an in silico alternative to traditional experimental approaches, potentially accelerating the development of DNAzymes.