Automatic reduction of false positive rates for RNA similarity searches in huge datasets

• Applicant: Zasha Weinberg, Ph.D.
• Subject Area: Bioinformatics and Theoretical Biology ; Biochemistry
• Term: since 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 393106201

Structured RNAs, which conserve a functionally important secondary structure, occur in all three domains of life, and include catalytic RNAs, metabolite-binding riboswitches and other regulatory RNAs. A fundamental task to further propel RNA research is to search sequence databases for structured RNAs of a given type. This task is called similarity search or homology search. If many homologous examples of a given RNA are known, covariance models (CMs) perform well in searches for structured RNAs. However, with ever-growing sequence databases, their false-positive rates are becoming problematic in many cases. Moreover, sometimes searches are conducted to find RNAs that are rare or are hypothesized variants of other known structured RNAs. In these cases, search methods based on pattern matching, which can exploit biochemical knowledge about the RNAs, are more appropriate. Unfortunately, these approaches often have even larger false-positive rates. Moreover, to reduce false positives, scientists must manually make their patterns more stringent—a time-consuming process that is difficult to perform well. Here, I propose multiple algorithms that, based on preliminary data, immensely reduce false-positive rates—e.g., by 10,000-fold or more. These algorithms are completely automatic, and thus do not burden scientists with extra work. They exploit information that is not currently used by CMs, so the methods are also expected to improve CMs’ false positive rates. The proposed project will develop these algorithms further and improve and quantify their performance. Then the improved search methods will be used to discover new forms of catalytic and metabolite-binding RNAs.

• DFG Programme: Research Grants