S

Saprot 650M AF2

Developed by westlake-repl
SaProt is a pre-trained model based on protein sequence and structural information, specifically optimized for low pLDDT regions.
Downloads 5,630
Release Time : 10/2/2023

Model Overview

SaProt is a pre-trained model that integrates protein sequence and structural information, capable of handling protein sequences containing low-confidence structural regions, supporting mutation effect prediction and protein embedding generation.

Model Features

Low pLDDT region handling
Capable of effectively processing low-confidence structural regions (pLDDT < 70) in protein sequences.
Mutation effect prediction
Supports predicting the effects of mutations at specific positions in protein sequences, including single-point and combinatorial mutations.
Protein embedding generation
Can generate embedding representations of protein sequences for downstream task analysis.
Two usage modes
Provides two usage pathways: via Huggingface interface and the original ESM method.

Model Capabilities

Protein sequence analysis
Mutation effect prediction
Protein embedding generation
Low-confidence region handling

Use Cases

Protein engineering
Mutation effect prediction
Predict the impact of mutations at specific positions on protein structure and function
Obtain effect values and probability distributions after mutation
Protein research
Protein representation learning
Generate embedding representations of protein sequences
Can be used for downstream tasks such as protein classification or function prediction
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase