RNALfold

What is RNALfold?

RNALfold computes locally stable RNA secondary structures using a sliding window approach. Part of the ViennaRNA Package, it identifies regions within long sequences that can form stable hairpins, stems, and other structural motifs without requiring the entire sequence to fold cooperatively.

The algorithm scales linearly with sequence length when the window size is fixed, using O(n + L²) memory and O(n · L²) CPU time where n is sequence length and L is window size. This efficiency makes RNALfold practical for scanning entire chromosomes or genomes to identify structured elements.

When to use RNALfold

RNALfold is designed for scenarios where global folding is inappropriate or computationally prohibitive:

• Long sequences: mRNAs, viral genomes, or chromosomal regions where global folding would be meaningless
• Local structure discovery: Finding hairpins, stem-loops, or regulatory elements embedded in longer contexts
• Genome annotation: Screening for structured non-coding RNAs like miRNA precursors or riboswitches

How does RNALfold work?

RNALfold applies the Zuker minimum free energy algorithm within a sliding window. For each position in the sequence, it considers only base pairs where both partners fall within a window of the specified size. This constraint prevents unrealistic long-range pairings that would never form in a biological context.

The algorithm reports locally optimal structures: those that cannot be improved by small perturbations within their region. A structure is reported only if it represents a genuine local minimum in the energy landscape, filtering out suboptimal folds that would be outcompeted by nearby alternatives.

Relationship to RNAplfold

While RNALfold returns discrete structures (in dot-bracket notation), RNAplfold computes base pair probabilities across the ensemble of possible local structures. RNALfold answers "what is the most stable local structure here?" while RNAplfold answers "how likely is each position to be paired?"

For most applications involving target site accessibility or structural diversity, RNAplfold provides more informative output. RNALfold is better suited when specific structural predictions are needed.

How to use RNALfold online

Paste an RNA sequence of any length, choose a window size, and RNALfold scans the full sequence for locally stable secondary structures. Each hit is returned with its dot-bracket notation, free energy, and exact start/end coordinates on the input sequence.

Input

Input	Description
RNA Sequences	One or more RNA sequences in FASTA format or plain text. Supports very long sequences (up to 50 MB).

Settings

Local folding options

Setting	Description
Window size	Maximum span of local structures (50-500, default 150). Larger values detect longer-range pairings but increase computation time.
Temperature	Folding temperature in degrees Celsius (0-100, default 37). Affects thermodynamic stability calculations.

Output

Results are returned as a spreadsheet with multiple rows per input sequence (one per local structure found):

Column	Description
Sequence ID	Identifier from FASTA header or auto-generated
Structure #	Index of the local structure within this sequence
Start Position	1-based start coordinate of the local structure on the input sequence
End Position	1-based end coordinate of the local structure on the input sequence
Window Span	Number of nucleotides between the start and end positions
Length	Length of the local structure in nucleotides
Local Structure	Dot-bracket notation showing the structure
Energy	Free energy of the local structure in kcal/mol

Choosing window size

The window size parameter determines the maximum span of detectable structures:

Window Size	Suitable For
50-100	Small hairpins, miRNA precursors
100-200	Typical regulatory elements, tRNAs
200-300	Larger riboswitches, some ribozymes
300-500	Complex structured domains

Smaller windows run faster and produce more focused results. Larger windows can detect more complex structures but may report many overlapping alternatives.

Limitations

RNALfold shares limitations common to ViennaRNA tools:

• No pseudoknots: The algorithm cannot predict structures with crossing base pairs
• Thermodynamic model only: Kinetic effects and co-transcriptional folding are not considered
• No tertiary interactions: Only secondary structure (base pairing) is predicted

For very long sequences, the number of reported structures can be large. The output is limited to 50 structures per sequence to keep results manageable.