Albumin Open-source Aptamer Affinity Prediction Model - Quickly Compare the Affinity Differences between 15-second Aptamers and Albumin

Albumin

Developed by Vilnius-Lithuania-iGEM

An aptamer affinity prediction model fine-tuned based on the Albert-base-v2 architecture, specifically designed to compare the affinity differences of 15-second-long aptamers with albumin.

Protein Model

Transformers

#Aptamer affinity prediction #Protein sequence classification #Biological sequence embedding

Downloads 24

Release Time : 3/2/2022

Model Overview

By learning the embedded positional representations of aptamers, this model can predict the affinity differences of various sequences to albumin, primarily used in the field of biological sequence analysis.

Model Features

Protein sequence feature learning

Fine-tuning enables the model to learn the embedded positional representations of aptamers, enhancing sequence discrimination capability.

High-precision prediction

Achieves approximately 90% accuracy in predicting aptamer affinity superiority.

Extensible architecture

Supports further fine-tuning on aptamers of different lengths or expanded datasets.

Model Capabilities

Masked aptamer sequence classification

Protein-aptamer affinity prediction

Biological sequence feature extraction

Use Cases

Biomedical research

Aptamer screening

Predicts the affinity differences of various aptamer sequences to albumin.

Accuracy 86.01%, F1 score 0.8618

Drug development assistance

Evaluates the binding capability of potential drug molecules to target proteins.

🚀 Albumin-15s

This model is a variant of Albert-base-v2, designed to compare 15 - nucleotide long aptamers and identify the one with higher affinity to the target protein, Albumin.

🚀 Quick Start

You can use this fine - tuned model for masked aptamer pair sequence classification, predicting which aptamer has a higher affinity for the target protein Albumin. It is also suitable for further fine - tuning on aptamers of different lengths or with expanded datasets.

✨ Features

The Albert model was initially pretrained in English. Due to the similarities between language, proteins, and aptamers, we fine - tuned it to help the model learn embedded positioning for aptamers, enabling better sequence discrimination.
More information can be found in our github and our iGEMs wiki.

📦 Installation

This section is not provided in the original README, so it is skipped.

💻 Usage Examples

Basic Usage

This model can be used to predict the relative affinity with a dataset preprocessing function that encodes specific data types (Sequence1, Sequence2, Label). The Label indicates binary information on whether Sequence1 has a higher affinity to the target protein Albumin.

from transformers import AutoTokenizer, BertModel
mname = "Vilnius-Lithuania-iGEM/Albumin"
model = BertModel.from_pretrained(mname)

Advanced Usage

To predict batches of sequences, you have to employ custom functions shown in git/prediction.ipynb.

📚 Documentation

Intended uses & limitations

The fine - tuned model can be used for masked aptamer pair sequence classification and prediction. However, its main purpose is to be fine - tuned again on aptamers of different lengths or with expanded datasets.

Limitations and bias

The fine - tuned Albert model for this task has a limitation of 90% accuracy in predicting which aptamer is more suitable for the target protein. Using Albert - large or an immense dataset of 15 - nucleotide aptamers could potentially increase the accuracy by a few percent. However, the extrapolation case has not been studied, and we cannot confirm that this model is state - of - the - art when one of the aptamers has extremely high affinity (almost maximum entropy to the Albumin).

🔧 Technical Details

This section is not provided in the original README, so it is skipped.

📄 Evaluation Results

Property	Details
Accuracy	0.8601
Precision	0.8515
Recall	0.8725
F1	0.8618
ROC AUC	0.9388

The score was calculated using sklearn.metrics.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご