P

Protst Esm1b

Developed by mila-intel
The ProtST framework enhances the pre-training and understanding of protein sequences through biomedical text, constructs the ProtDescribe dataset, designs three pre-training tasks, and supports supervised learning and zero-shot prediction.
Downloads 173
Release Time : 1/2/2024

Model Overview

ProtST is a pre-training framework that combines biomedical text and protein sequences, aiming to enhance the functional understanding and representation ability of protein language models.

Model Features

Multimodal pre-training
Pre-train by combining protein sequences and biomedical text to enhance the model's understanding of protein functions.
Zero-shot prediction
Supports zero-shot classification tasks and can perform protein function prediction without additional training.
High-performance representation learning
Outperforms existing models on various representation learning benchmarks, especially in function prediction tasks.

Model Capabilities

Protein sequence representation learning
Protein function prediction
Zero-shot classification
Multimodal alignment

Use Cases

Biomedical research
Protein subcellular localization prediction
Predict the localization of proteins in cells, such as the nucleus, mitochondria, etc.
Shows superior performance in zero-shot tasks.
Protein function annotation
Automatically add functional descriptions to protein sequences.
Improves annotation accuracy through multimodal alignment.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase