Cart 2

RBS Strength Prediction & Design Tool (InceptRBS)

This tool uses a deep learning model (Inception-CNN) to predict translation strength, explain key positions, and generate new sequences for bacterial (E. coli) ribosome binding site (RBS) DNA sequences.

Core functions:

1. Predict — Input a 17 bp DNA sequence to predict its RBS translation strength (0-1) and percentile ranking in the full training dataset.

2. Explain — Based on the ISM (In Silico Mutagenesis) algorithm, mutate the sequence position by position and calculate each base's contribution to translation strength to identify key functional regions.

3. Generate — Based on the Hill Climbing algorithm, automatically design high-strength, low-strength, or target-strength RBS sequences for synthetic biology experiment design.

1. RBS Task Parameters:

Parsed sequences: 0, total bases: 0

DNA sequence input (for explain mode, up to 100 FASTA entries):



Usage

  • RBS strength prediction: the model only handles 17 nt DNA sequences (17 nt upstream of the start codon), e.g. CTCGAGGAGATATACAT[ATG], so you should input CTCGAGGAGATATACAT. If the input length is greater than 17 nt, it will be truncated to 17 nt from the 5' end. If shorter than 17 nt, it will be padded with 'N' at the end to length 17 nt. For best accuracy, manually provide a 17 nt sequence.
  • RBS design has two modes: 1) with sequence constraints: only one sequence is supported; replace bases to be changed with 'N' in the 17 nt template. 2) without sequence constraints: no input sequence is required; the model will automatically design 10 RBS sequences.
  • The model outputs translation strength (0-1) and percentile ranking in the full training dataset.
  • The score of the candidate key region indicates how much it explains sequence importance (0-1).

Model Performance Metrics

---------------------------------------------------
Metric                  Train Set   Test Set
---------------------------------------------------
R²                      0.945       0.911
MAE                     0.039       0.048
MSE                     0.004       0.006
---------------------------------------------------

Last updated: 2026-05-05

References

Höllerer S, Papaxanthos L, Gumpinger AC, Fischer K, Beisel C, Borgwardt K, Benenson Y, Jeschek M. Large-scale DNA-based phenotypic recording and deep learning enable highly accurate sequence-function mapping. Nat Commun. 2020 Jul 15;11(1):3551. doi: 10.1038/s41467-020-17222-4. PMID: 32669542; PMCID: PMC7363850.