santiagoahl / rna-taxonomy-prediction

We worked with an open csv-dataset which consist on RNA sequences with several taxonomies. Using python we were able to create an XGBoost model that classifies that sequence into 1 of 19 differents taxonomies. We also worked with Markov chains in order to treat the data.

RNA Taxonomy Classification

An XGBoost Multiclass classifier built in scikit-learn using Markov Chains.

Key Features • How To Use • Credits • License

Key Features

• This machine learning model takes a RNA sequence and predicts what class does it belong to. Classes are taken as taxonomies. The avaible taxonomies are the following 19:

  • Orthomyxoviridae
  • Rhabdoviridae
  • Arteriviridae
  • Coronaviridae
  • Reoviridae
  • Caliciviridae
  • Phenuiviridae
  • Hantaviridae
  • Picornaviridae
  • Betaflexiviridae
  • Astroviridae
  • Closteroviridae
  • Flaviviridae
  • Potyviridae
  • Retroviridae
  • Togaviridae
  • Paramyxoviridae
  • Hepeviridae
  • Pneumoviridae

• Before Prediction the model computes a markov chain whose states are the 64 writeable codons with the nucleoids A, C, G, T and then computes metrics over its adjacent associated matrix: 8 of them are matricial norms and the missing 10 parameters are the first eigenvalues complex norms ascending ordered. Namely:

  • Frobenius Norm
  • Nuclear Norm
  • Infty Norm
  • Neg Infty Norm
  • Neg L1 Norm
  • L1 Norm
  • Neg L2 Norm
  • L2 Norm
  • eig 1
  • eig 2
  • eig 3
  • eig 4
  • eig 5
  • eig 6
  • eig 7
  • eig 8
  • eig 9
  • eig 10

With these new metrics, we built a new dataset. and we found this scatter plot:

• We implemented a Random Forest model whose train data is taken from the new dataset.
• We archieved a 96.9% of F1 score on validation set.
• The confusion matrix is the following

• The learning curve is the following

How To Use

To clone and run this application, follow these steps

# Clone this repository
$ git clone https://github.com/santiagoahl/rna-taxonomy-prediction.git

# Go into the repository
$ cd rna-taxonomy-prediction

# Go to jupyter notebooks
$ jupyter-notebook

# Run the Libraries & Modules cell
# Run the Model Import cell

Credits

This software uses the following packages:

• Scikit-Learn
• Numpy
• Joblib

License

MIT

---

Web Site santiagoal.super.site  ·  GitHub @santiagoahl  ·  Twitter @sahumadaloz