IMPACT: Interpretable Microbial Phenotype Analysis via microbial Characteristic Traits
We present a framework that combines a feature engineering step to transform tabular ASV abundance data to sample wised image-format using functional microbial annotation databases, with a residual spatial attention transformer block architecture for phenotype classification.
We find that our model delivers improved predictive accuracy performance across multiclass classification compared to similar methods. More importantly, our approach provides interpretable feature importance through image classification saliency methods. This enables the extraction of taxa markers (features) associated with a disease outcome but also their associated functional microbial traits and metabolites.
To correctly use IMPACT via your local device, we suggest first create a conda environment by:
conda create -n <env> python=3.9
conda activate <env>After entering an entire new virtue environment, the recommended way to install our package is:
pip install IMPACTBIOINFORMATICSThis commend would automatically download and install the lastest version (should be after 4.7) onto your local device.
Another alternative way to install the package is using the .tar file provided in this repo by:
pip install IMPACTBIOINFORMATICS-X.X.tar.gzAfter this, validate the installation success in python by:
import impactIf no error occurs, then IMPACT has been sucessfully installed.
There are two main functions implemented within IMPACT, "train()" and "predict()".
Function train(input_file, metabolites_file, isize, check_feature_importance) could help user to train a Phenotype Prediction Model and analysis the data via microbial characteristic traits. It has 4 arguements, respectively are:
--input_file: should be a str. The absolute path of the input file,
should be a .csv file, containing a matirx of normalized data.
Each line represents a patient while each column represent a type of bacteria.
Note that the last line is phylum information of all measured bacterias.
--metabolites_file: should be a str. The absolute path of the metabolites file,
should be a .csv file, containing a matirx of normalized data.
Each line represents a type of bacteria while each column represent a kind of metabolite.
IF you DO NOT have your customized one, please use the file provided in this repo.
--isize: should be an int. Default value is set to 60, determined by
hyper-parameter searching via a couple cross validation,
This arguement represents the image size of transformed input data.
--check_feature_importance: should be a bool. Default is set as Ture.
This is the switch to draw the saliency plot.
Important for microbial characteristic trait analysis.
If True, would produce a .png file like following.
Function predict(fname,input_file, isize) could help user to predict patient phenotype via a pre-trained Prediction Model (details could be found in the description above). It has 3 arguements, respectively are:
--fname: should be a str. Default is set as "model.pth".
This is the name of the pre-trained model.
Please find the .pth file after your training by Function "train()".
Use the name of this file.
--input_file: should be a str. The absolute path of the input file,
should be a .csv file, containing a matirx of normalized data.
Each line represents a patient while each column represent a type of bacteria.
Note that the last line is phylum information of all measured bacterias.
--isize: should be an int. Default value is set to 60, determined by
hyper-parameter searching via a couple cross validation,
This arguement represents the image size of transformed input data.
If you find our codes useful, please consider citing our work:
@article{IMPACT,
title={IMPACT: Interpretable Microbial Phenotype Analysis via microbial Characteristic Traits},
author={},
journal={},
year={2024},
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent