sparkjj/aa2dna-1000

Amino Acid → DNA Back-Translation Dataset

Description

This dataset contains short amino acid sequences and one possible DNA coding sequence for each of them. All examples are synthetic and were generated using the standard genetic code by back-translating random protein sequences. The dataset is intended as a simple sequence-to-sequence toy task where a model learns to map amino acid sequences back to plausible coding sequences.

The data are not taken from any real organism and do not correspond to known genes or proteins. Codons for each amino acid are chosen uniformly at random among the standard genetic-code options, so codon usage, expression level, or species-specific bias are not modeled. The dataset is suitable for demonstrations and teaching with the Hugging Face ecosystem, but it should not be used directly for biological design or analysis.

Dataset Structure

Each example is a pair of strings:

• src: input amino acid sequence (length 5–20, using the 20 standard amino-acid letters)
• tgt: DNA coding sequence obtained by back-translation (length 15–60, multiple of 3, composed of A/C/G/T)

All characters are ASCII, and no stop codons are included in the amino acid sequences. Every DNA sequence in the dataset is compatible with the corresponding amino acid sequence under the standard genetic code.

The dataset is split into three parts:

• train: 800 examples
• validation: 100 examples
• test: 100 examples

All splits share the same (src, tgt) schema, and duplicate pairs have been removed so that each example is unique.