Beta-Lactamase Fitness Prediction Dataset

This dataset is designed for training and evaluating machine learning models on protein fitness landscape prediction. It specifically focuses on the beta-lactamase enzyme, a primary driver of antibiotic resistance in bacteria.

Dataset Overview

The primary objective of this dataset is to map protein sequences to their functional activity. In the context of beta-lactamase, this functional activity—often referred to as fitness—typically represents the enzyme's ability to hydrolyze beta-lactam antibiotics.

Predicting these values computationally allows for the rapid screening of protein variants without the need for exhaustive wet-lab experimentation.

Task Definition

This is a supervised regression task. Given a primary amino acid sequence as input, the model must predict a continuous numerical value representing the fitness or deep mutational scanning (DMS) score of that variant.

• Input: A string of amino acids representing a mutated version of the wild-type beta-lactamase protein.
• Output: A numerical fitness score. Higher values generally indicate higher enzymatic activity or increased antibiotic resistance.

Data Structure

The dataset consists of training and testing splits provided in a standard tabular format. Each entry contains:

• sequence: The full primary protein sequence.
• target: The normalized fitness score or experimental measurement (e.g., log-enrichment or MIC values).

Usage

The dataset is compatible with the Hugging Face datasets library. It can be loaded and integrated into training pipelines as follows:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("hazemessam/beta-lactamase")

# Access a sample
sample = dataset['train'][0]
print(f"Sequence: {sample['sequence']}")
print(f"Fitness Score: {sample['target']}")