Important Note

• Each of our jupyter notebook takes about an hour to run due to

  • the function to choose number of clusters (pairwiseCrossValidation) takes about 20 minutes to run
  • the function to run the alignment algorithm (runClustalInRange) takes about 30 minutes to run

• DNABERT6 and Finetuned_DNABERT6 takes about 4-6 hours to get the embeddings

For these reasons we provide all the necessary files to get our results, in this drive link: https://drive.google.com/drive/folders/1KLiCzlLoEf0avWA5S5f40Lqq-E_cfNEX?usp=sharing

Repository Organization

• run.ipynb : this is the model with the best results, the training loop is commented, you can just load the .pth file.

  • We recommend not to run "runClustalInRange" function, you can find the required files in our drive folder mentioned above.

• helper.py : we have all our utility functions in this file, all notebook import this file

• plot_clusters.py : this function was provided to us by Antoine Tappy who is working on a similar project

• data_preprosessing.ipynb : we did all our data preprocessing in this file, you can find all the data we used in Data.zip

• Other_Notebook : as we tried many different models, we keep all of our other notebooks here, there are some duplicate files such as helper.py, to simplify importing

Requirements

• clustal
• biopython
• numpy
• sklearn
• scipy
• pytorch
• matplotlib

Data & Pretrained Models:

Data

We got our data from NCBI's website

The procedure is as follows

• Enter rbcL
• Filter for plants only
• Select the sequence length
  • rbcL : 600 to 1000 -> ~90k Sequence, ~80Mo
• Download it to have the information and the DNA sequence Click and send to (corner top right) > Complete Record > File > Format = Fasta > Sort by Taxonomy ID
• Put the fasta file into /Data

Pretrained models

• We used 2 pretrained models

  • We used models from DNABERT
    • Pretrained DNABERT6
    • FineTuned DNABERT6

• We also provide our trained models as .pth files in our repository

You can find these models in our drive folder as well

Running run.ipynb

• To be able to run run.ipynb, please extract Data.zip with the same name in the same directory as run.ipynb
• As mentioned before in "Important Note" we highly recommend downloading /clustal /clusters and /plots folders from our drive to avoid running the notebook for an hour