Recipient independent high accuracy FMT prediction and optimization

This code is attached to the paper "Recipient independent high accuracy FMT prediction and optimization". We propose a tool to predict the recipient phenotype a week after the FMT using only the donors' microbiome (iMic). We then extend the method to optimize the best-planned transplant (Bacterial cocktails) by combining the predictor and a Genetic Algorithm (GA).

iMic - prediction

iMic predicts a recipient's phenotype t days after the FMT, where t is a parameter of iMic. iMic is based on the donor's properties only. For more details see the paper.

How to apply iMic

1. Add the dataset to the nni_data_loader function. We recommend to preprocess the data via the MIPMLP pipeline with the following parameters Taxonomy Level = Specie, Taxnomy Group = Mean, Normalization = Log. The MIPMLP pipeline can be found at MIPMLP website.

2. The rest parameters for the function load_nni_data are: name_of_dataset = 'mean', and D_mode = 'dendogram'.

3. Use the micro2matrix code to generate the images from the preprocessed microbiome.

4. Use main_nni_runner to apply iMic on the images, the tag can be any recipient's phenotype (such as: Shannon alpha diversity, orders relative abundances, etc. Just change the a_div variable to the target outcome. The parmaters that should be used are:

   sys[0] = mean (name_of_dataset)

   sys[1] = dendogram (D_mode)

   sys[2] = cnn2 (model)

   Notice - iMic can also be appied separately via the following pypi

Genetic Algorithm (GA) - optimization

To find the optimal required FMT given a targeted outcome, such as: maximizing the recipients' Shannon index, minimizing the recipients' Shannon index, or maximizing the relative abundances of a certain order in the recipients' samples, we developed a Genetic Algorithm (GA). For more details see the paper.

How to apply Genetic Algorithm

All the relevant code for the Shannon index task can be found at Genetic_algorithm.py.

1. Prepare a pre-trained iMic model that would be as the oracale of the GA.

2. Define the fitness function, in our case, there is a fitness function of maximizimg or minimizing the recipient's Shannon index a week after FMT while controlling the number of non-zero taxa in the FMT (gamma). If gamma is 0, it means all the taxa can be used.

General tools

There is a function for merging different datasets at tools.py.

The Shannon index can be calculated at this colab.

Running a trained version of iMic

Note there is an option to run a trained version of iMic on the Shannon of the human to human cohort. By getting the weights from this Drive and uploading the files of the mapping, the image of donors, the donors' Shannon and outcome data. The code for applying the model can be found at "run_trained_model.py".

Contact us

Oshrit Shtossel - oshritvig@gmail.com