Biofit is a machine learning library designed for bioinformatics datasets. It provides tools for transforming, extracting, training, and evaluating machine learning models on biomedical data. It also provides automatic data preprocessing, visualization, and configurable processing pipelines. Here are some of the main features of Biofit:
- Automatic Data Preprocessing: Automatically preprocess biomedical datasets using built-in preprocessing steps.
- Automatic Visualization: Automatically visualize data using built-in visualization methods geared towards biomedical data.
- Configurable Processing Pipelines: Define and customize data processing pipelines.
- Data Handling Flexibility: Support for a wide range of data formats, including:
- Machine Learning Models: Supports a wide range of machine learning models, including:
- Caching and Reuse: Caches intermediate results using Apache Arrow for efficient reuse.
- Batch Processing and Multiprocessing: Utilize batch processing and multiprocessing for efficient handling of large-scale data.
You can install Biofit via pip:
pip install biofitBiofit provides preprocessing capabilities tailored for omics data. You can use built-in classes to load preprocessing steps based on the experiment type or create custom preprocessing pipelines. The preprocessing pipeline in Biofit uses a syntax similar to sklearn and supports distributed processing.
Biofit allows you to fit and transform your data in a few lines, similar to sklearn. For example, you can use the LogTransformer to apply a log transformation to your data:
from biofit.preprocessing import LogTransformer
import pandas as pd
dataset = pd.DataFrame({"feature1": [1, 2, 3, 4, 5]})
log_transformer = LogTransformer()
preprocessed_data = log_transformer.fit_transform(dataset)
# Applying log transformation: 100%|█████████████████████████████| 5/5 [00:00<00:00, 7656.63 examples/s]
print(preprocessed_data)
# feature1
# 0 0.000000
# 1 0.693147
# 2 1.098612
# 3 1.386294
# 4 1.609438You can automatically apply standard preprocessing steps by specifying the experiment type. This allows you to load tailored preprocessing steps for the type of data you are working with, such as "otu", "asv", "snp", or "maldi":
from biofit.preprocessing import AutoPreprocessor
preprocessor = AutoPreprocessor.for_experiment("snp", [{"min_prevalence": 0.1}, None])
print(preprocessor)
# [('min_prevalence_row', MinPrevalencFilter(min_prevalence=0.1)),
# ('min_prevalence', MinPrevalenceFeatureSelector(min_prevalence=0.01))]
# Fit and transform the dataset using the preprocessor
preprocessed_data = preprocessor.fit_transform(dataset)Biofit is made with Biosets in mind. You can pass the loaded dataset instead of a string to load the preprocessors:
from biosets import load_dataset
dataset = load_dataset("csv", data_files="my_file.csv", experiment_type="snp")
preprocessor = AutoPreprocessor.for_experiment(dataset)
print(preprocessor)
# [('min_prevalence_row', MinPrevalencFilter(min_prevalence=0.01)),
# ('min_prevalence', MinPrevalenceFeatureSelector(min_prevalence=0.01))]
preprocessed_data = preprocessor.fit_transform(dataset)Biofit allows you to create custom preprocessing pipelines using the
PreprocessorPipeline class. This allows chaining multiple preprocessing steps from
sklearn and Biofit in a single operation:
from biofit import load_dataset
from biofit.preprocessing import LogTransformer, PreprocessorPipeline
from sklearn.preprocessing import StandardScaler
# Load the dataset
dataset = load_dataset("csv", data_files="my_file.csv")
# Define a custom preprocessing pipeline
pipeline = PreprocessorPipeline(
[("scaler", StandardScaler()), ("log_transformer", LogTransformer())]
)
# Fit and transform the dataset using the pipeline
preprocessed_data = pipeline.fit_transform(dataset.to_pandas())For further details, check the advance usage documentation.
Biofit is licensed under the Apache 2.0 License. See the LICENSE file for more information.
If you would like to contribute to Biofit, please read the CONTRIBUTING guidelines.