BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
Paper β’ 2301.12597 β’ Published β’ 3
GeneLinguaLM is a multimodal model that generates natural language descriptions of protein functions from amino acid sequences.
GeneLinguaLM bridges protein sequences and natural language through cross-modal learning:
| Model | ROUGE-1 | ROUGE-L | BLEU |
|---|---|---|---|
| GeneLinguaLM v5 | 0.2295 | 0.1561 | 0.0315 |
| Zero-shot Mistral-7B | 0.1709 | 0.1160 | 0.0075 |
| BioGPT | 0.1513 | 0.1092 | 0.0063 |
| Mol-Instructions | 0.0025 | 0.0025 | 0.0007 |
GeneLinguaLM outperforms:
from genelinguaLM import GeneLinguaLM
# Load model
model = GeneLinguaLM()
# Describe a protein sequence
sequence = "MALWMRLLPLLALLALWGPDPAAAFVNQHLCGSHLVEALYLVCGERGFFYTPKTRREAEDLQVGQVELGGGPGAGSLQPLALEGSLQKRGIVEQCCTSICSLYQLENYCN"
description = model.describe(sequence)
print(description)
# Output: "Hormone that regulates glucose metabolism and blood sugar levels..."
Input: Human Insulin sequence
MALWMRLLPLLALLALWGPDPAAAFVNQHLCGSHLVEALYLVCGERGFFYTPKT...
Output:
Hormone that regulates carbohydrate and lipid metabolism.
Plays a key role in the regulation of glucose levels in the blood...
Protein Sequence
β
[ProtBERT] β Sequence Embeddings (1024-dim)
β
[Q-Former] β 32 Query Tokens (768-dim)
β
[Projector] β LLM Embeddings (4096-dim)
β
[Mistral-7B + LoRA] β Natural Language Description
checkpoint_step15732.pt: Main checkpoint (Q-Former + Projector weights)lora_step15732/: LoRA adapter for Mistral-7Bqformer_checkpoint.pt: Q-Former pretrained weightstorch>=2.0
transformers>=4.35
peft>=0.6
@misc{genelinguaLM2024,
title={GeneLinguaLM: Bridging Protein Sequences and Natural Language},
author={GeneLinguaLM Team},
year={2024},
url={https://github.com/powersimmani/geneLLM}
}
Apache 2.0
Base model
Rostlab/prot_bert_bfd