EVA: Benchmark and Experiment Data for RNA Generation and Codon Optimization

This dataset accompanies the EVA (Evolutionary Versatile Architect) paper and contains all benchmark and experiment data for RNA sequence generation and codon optimization tasks. EVA is a 1.4-billion-parameter RNA foundation model capable of species-conditioned, RNA-type-conditioned, and unconditional RNA generation, as well as codon optimization via directed evolution. The dataset is organized into four archives: 1. mrna_codon_optimization.tar.gz mRNA and circular RNA (circRNA) codon optimization results from EVA and baseline models (CodonFM-1B, Evo2-1B, GemoRNA, random synonymous substitution). mRNA targets: linear mRNA (8 CDS variants), HIV-1 gp160, PR8 HA, RABV-G, VZV gE circRNA targets: CAR-T, RABV-G, SARS-CoV-2 (Group-I intron), SARS-CoV-2 (T4 ligase) Each target contains input wild-type sequences, EVA-optimized sequences (with and without Homo sapiens lineage conditioning), and baseline model outputs. All in FASTA format. 2. mrna_6species.tar.gz EVA-generated and natural mRNA sequences for 6 species, benchmarking species-conditioned generation. Species: Homo sapiens, Mus musculus, Rattus norvegicus, Cricetulus griseus, Drosophila melanogaster, Caenorhabditis elegans 1,000 generated mRNA sequences per species (FASTA) and natural reference sequences Metrics: MFE, GC content, k-mer frequency profiles, loop/helix proportions, base-pairing percentages 3. generrna_generation.tar.gz EVA-generated rRNA sequences benchmarked against the GenerRNA database (unconditional generation). 100,000 natural rRNA sequences from GenerRNA (reference) and 1,000-sequence subsample EVA-generated rRNA-like sequences Metrics: k-mer frequency profiles, MFE, GC content, loop/helix proportions, base-pairing percentages 4. generated_rna_sequences.tar.gz EVA-generated RNA sequences spanning 15 RNA biotypes with matched natural controls, benchmarking multi-type controllable generation. RNA biotypes (11): mRNA, lncRNA, circRNA, tRNA, rRNA, miRNA, piRNA, sRNA, snRNA, snoRNA, viral RNA ~200 generated sequences per type (FASTA) and up to 5,000 natural reference sequences per type Metrics: MFE, GC content, k-mer frequency profiles, loop/helix proportions, base-pairing percentages All sequence files are in FASTA format. Metric files are in CSV or NumPy .npz format. This dataset is associated with the EVA paper (preprint forthcoming).Code: https://github.com/yanjiehuang/EVA1