RAPPTOR prediction
Prediction help
Configure the prediction sequence, reference-genome context, and model parameters, then interpret the resulting scores and promoter calls.
Prediction workflow
- 1Provide the sequence for prediction
An input of exactly 100 bp is evaluated once per selected strand. Longer inputs are scanned as overlapping 100 bp windows. Inputs shorter than 100 bp are not supported.
- 2Select the complete reference genome
Select the genome from which the input sequence was obtained. The reference genome provides the genomic context used during prediction.
Genome context and CGR
CGR (Chaos Game Representation) encodes DNA sequence patterns across the selected complete reference genome. RAPPTOR uses this representation as genomic context when evaluating the Step 1 sequence. The CGR is generated automatically; no CGR file is required.
Select the same strain and assembly version from which the input sequence was obtained. A different reference genome provides a different genomic context and may alter the prediction result.
- 3Configure prediction parameters
Select the strand mode and, for sequence scans, the promoter cutoff and stride.
- Strands
- Select Both when sequence orientation is unknown. Forward evaluates the submitted orientation; Reverse evaluates its reverse complement.
- Promoter cutoff
- Defines the minimum score considered during promoter reporting. The default is appropriate for standard predictions.
- Stride
- Defines the distance between consecutive 100 bp windows. A stride of 1 provides base-level sampling; larger strides sample fewer positions.
Sequence and reference requirements
Raw DNA and FASTA input are supported. Reference files may use .fa, .fasta, or .fna, optionally gzip-compressed. Standard IUPAC DNA characters are accepted; U is interpreted as T and ambiguous bases as N.
Parameter behavior
The cutoff affects promoter reporting, not the underlying model scores. Increasing stride accelerates a scan but reduces positional resolution because fewer windows are evaluated.
Interpreting prediction results
- Model score
- Higher scores indicate that a 100 bp window is more promoter-like for the selected model and genomic context. A model score is neither a probability nor experimental evidence.
- Promoter call
- At stride 1, a reported promoter must be a local maximum above the selected cutoff. A score above the cutoff is therefore not sufficient by itself to produce a promoter call.
- Score track
- BigWig tracks retain the complete set of computed scores, including values below the promoter cutoff.
- Promoter track
- GFF3 records contain the genomic positions, strands, and scores of reported promoter predictions.
Interpretation boundary
RAPPTOR results are computational predictions. They do not establish experimental promoter activity or identify a transcription start site.
How promoter calls are generated
For stride 1, scores are Gaussian-smoothed by strand before local maxima are identified and filtered by the promoter cutoff. For larger strides, each sampled raw-score window above the cutoff is reported directly.
Result files and retention
Use GFF3 for discrete promoter intervals and BigWig for continuous score visualization in a genome browser. Result artifacts are temporary and should be downloaded before the expiry time shown on the task page.
Troubleshooting
Reference genome not found
Search with a versioned GCA_ or GCF_ accession. If the assembly is unavailable, upload its complete-genome FASTA.
FASTA input rejected
Verify the file extension, confirm that records are non-empty, and check the displayed size limit. Sequences shorter than 100 bp cannot be evaluated.
No promoters reported
Inspect the score tracks before changing the cutoff. At stride 1, windows above the cutoff must also be local maxima to be reported as promoters.
Task failed or results expired
Confirm the prediction sequence and reference genome, then submit a new task. Expired result artifacts cannot be recovered from the task page.