yangheng/OmniGenome-52M

Multi-species Foundation Model for Universal RNA and DNA Downstream Tasks

Notes

We are keep updating the checkpoints, the current checkpoint is trained for 0.85 epoch.

Training Examples

Refer to GitHub https://github.com/yangheng95/OmniGenome

Usage

This model is available for replacing genomic foundation models such as CDSBERT, Nucleotide Transformers, DNABERT2, etc.

from transformers import AutoModel
model = AutoModel.from_pretrained("yangheng/OmniGenome-52M", trust_remote_code=True)

Subtasks

• Secondary structure prediction
• Genome Sequence Classification
• Genome Sequence Regression
• Single Nucleotide Repair
• Genome Masked Language Modeling
• etc.

Part of the codes are adapted from ESM2.