---
license: mit
language:
- en
tags:
- rna
- biology
- diffusion
- generative
- sequence-design
---

# yakRNA Design: A semantic multimodal RNA composer

<p align="center">
  <img src="https://raw.githubusercontent.com/YousufAKhan/yakRNA/main/mascot.png" width="160"/>
</p>

**yakRNA Design** is a 110M-parameter discrete diffusion language model for conditional RNA sequence design. It can generate RNA sequences conditioned on any combination of secondary structure, consensus sequence, and Gene Ontology (GO) terms — or unconditionally from a target length.

---

## Model Details

| | |
|---|---|
| **Architecture** | ModernBERT-based discrete diffusion |
| **Parameters** | 110M |
| **Training data** | Full Rfam database |
| **Max sequence length** | 636 nt |
| **Supported GO terms** | 280 |

---

## Quickstart

The easiest way to use yakRNA Design is via the **[Google Colab notebook](https://colab.research.google.com/github/YousufAKhan/yakRNA/blob/main/yakRNA_colab.ipynb)** — no setup required.

For local use, see the **[GitHub repository](https://github.com/YousufAKhan/yakRNA)**.

---

## Download the Weights

**CLI:**
```bash
pip install huggingface_hub
huggingface-cli download MasterYster/yakRNA-Design yakRNA_110M.pt --local-dir checkpoints/
```

**Python:**
```python
from huggingface_hub import hf_hub_download
hf_hub_download(repo_id="MasterYster/yakRNA-Design", filename="yakRNA_110M.pt", local_dir="checkpoints/")
```

---

## Generation Modes

| Mode | Description |
|------|-------------|
| Unconditional | Generate from a target length |
| Structure-conditioned | Dot-bracket secondary structure |
| Consensus-conditioned | Mixed-case IUPAC consensus |
| GO term-conditioned | Up to 12 GO terms |
| Sequence infilling | Fixed positions + `*` masks |
| Multimodal | Any combination of the above |

---

## Example Usage

```bash
# Install
git clone https://github.com/YousufAKhan/yakRNA.git
cd yakRNA
pip install -r requirements.txt

# Generate 5 sequences conditioned on a hairpin structure
python inference/rna_sequence_generator.py \
    --config configs/inference.yaml \
    --checkpoint checkpoints/yakRNA_110M.pt \
    --secondary_structure "((((....))))" \
    --num_sequences 5

# Generate with all three modalities
python inference/rna_sequence_generator.py \
    --config configs/inference.yaml \
    --checkpoint checkpoints/yakRNA_110M.pt \
    --secondary_structure ":::::::<<<<<<<<<-:::--[[[[[-->>>>>>>>><<<<<<<<<<_________>>>->>>>>>>::::]]]]]::::" \
    --consensus "GAGUaaGGGGuuCuAGU...gcaGCcCgcCUaGaaCCCUGcgacacuGGuucuaaaaCagAugucgUuuuaAGgGCuUUUG" \
    --go_terms "GO:0075523" \
    --num_sequences 5
```

---

## Links

- **GitHub**: [YousufAKhan/yakRNA](https://github.com/YousufAKhan/yakRNA)
- **Colab**: [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/YousufAKhan/yakRNA/blob/main/yakRNA_colab.ipynb)

---

## Citation

```bibtex
@software{yakrna2026,
  author = {Khan, Yousuf},
  title  = {yakRNA Design: A semantic multimodal RNA composer},
  year   = {2026},
  url    = {https://github.com/YousufAKhan/yakRNA}
}
```