chunribu / miidl

A Python package for biomarkers identification powered by interpretable deep learning

MIIDL

MIIDL /ˈmaɪdəl/ is a Python package for microbial biomarkers identification powered by interpretable deep learning.

---

Getting Started

👋Welcome!

🔗This guide will provide you with a specific example that using miidl to detect microbial biomarkers of colorectal cancer and predict clinical outcomes.

After that, you will learn how to use this tool properly.

---

Installation

pip install miidl

or

conda install miidl captum -c pytorch -c conda-forge -c bioconda

---

Features

• One-stop profiling
• Multiple strategies for biological data
• More interpretable, not a "black box"

---

Workflow

1) Quality Control

The very first procedure is filtering features according to a threshold of observation (non-missing) rate (0.3 by default).

2) Normalization

miidl offers plenty of normalization methods to transform data and make samples more comparable.

3) Imputation

By default, this step is inactivated, as miidl is designed to solve problems including sparseness. But imputation can be useful in some cases. Commonly used methods are available if needed.

4) Reshape

The pre-processed data also need to be zero-completed to a certain length, so that a CNN model can be applied.

5) Modeling

A CNN classifier is trained for discrimination. PyTorch is needed.

6) Interpretation

Captum is dedicated to model interpretability for PyTorch. This step depends heavily on captum.

---

Contact

If you have further thoughts or queries, please feel free to email at [e-mail] or open an issue!

---

Citation

@misc{jiang2021miidl,
      title={MIIDL: a Python package for microbial biomarkers identification powered by interpretable deep learning}, 
      author={Jian Jiang},
      year={2021},
      eprint={2109.12204},
      archivePrefix={arXiv},
      primaryClass={q-bio.QM}
}

---

License

MIIDL is released under the MIT license.