RNAflash: Fast and Sensitive Structure-aware RNA Homology Detection and Alignment

In the vast library of life, RNA acts as a versatile messenger and manager, carrying instructions from DNA and helping to build the proteins that run our cells. To understand how a specific piece of RNA works, scientists often look for its relatives—other RNA molecules that share a common ancestor. These relatives, called homologs, are like distant cousins in a family tree; they may look quite different on the surface, but they often fold into similar shapes to perform the same job. Finding these cousins is crucial for predicting how RNA folds into its three-dimensional structure, a key step in understanding how it functions and how it might be targeted by medicines. However, there is a persistent problem in this field: the tools used to find these relatives usually force scientists to choose between speed and accuracy. Fast tools often miss the distant cousins because they rely too heavily on exact letter-for-letter matches, while the accurate tools that look at the shape of the molecule are so slow they can take days to search through modern databases containing millions of sequences.

A team of researchers at Shanghai Jiao Tong University has introduced a new approach called RNAflash that aims to break this trade-off. Their method acts like a two-step search engine designed to find these elusive RNA relatives quickly and then arrange them into a useful family portrait. The first step uses a system trained to recognize the "voice" of RNA sequences, even when the letters have changed significantly over time. Instead of comparing every single letter of a query sequence against every sequence in a massive database of thirty million entries, this system converts each sequence into a compact summary. It then uses a high-speed index to find the most promising matches in a fraction of a second. This initial sweep is fast, but it is not perfect; it might include some false alarms or miss subtle connections that only a closer look would reveal.

To fix this, the second step of the process takes the top candidates from the first round and examines them with a much more careful eye. This stage uses a different kind of analysis that looks at the predicted shape of the RNA molecules. Since the shape of an RNA molecule is often more stable and conserved than its sequence of letters, this step can confirm relationships that the fast search missed. It aligns the sequences based on how their shapes match up, ensuring that the final list of relatives is both accurate and biologically meaningful. The result is a collection of sequences arranged in a multiple sequence alignment, which is a grid that lines up the relatives so scientists can see where they agree and where they differ. This grid is the foundation for predicting the molecule's 3D structure.

The researchers tested their system against a wide range of existing tools, including some that are considered the gold standard in the field. They found that RNAflash was able to find more distant relatives than the fast, sequence-only tools, and it did so much faster than the slow, shape-aware tools. In one large-scale test involving hundreds of queries against a database of thirty million sequences, the new method completed the entire search and alignment process for all 174 queries in 1,682 seconds. In contrast, the traditional shape-based tools took over 100 times longer to do the same job. The alignments produced by RNAflash were not just faster to create; they were also of higher quality, containing more useful evolutionary signals that help scientists understand how the RNA folds.

The team also created a lighter, faster version of their tool for situations where speed is the absolute priority. This version, called RNAflash-L, can find and align relatives in less than a second per query. While it is slightly less sensitive than the full version, it still outperforms other fast tools in the quality of the alignments it produces. This speed is significant because it means that researchers can now generate the high-quality family alignments needed for advanced structure prediction on a scale that was previously impossible. By making it possible to search through millions of sequences in seconds rather than days, this work opens the door to analyzing RNA molecules that were previously too difficult to study, potentially accelerating the discovery of new biological functions and therapeutic targets.