DNA-Barcode-Classification

B.E. Final Year Research Project 2020-2021

Description

Classification of an unknown specimen to its respective Taxonomy rank by analyzing its DNA Barcode.

Input

1. DNA Barcode as a raw FASTA sequence (string).

2. Choice of Supervised Machine Learning Model to use for prediction:

   • Naïve Bayes
   • Support vector machine
   • Random forest
   • k-nearest neighbors algorithm

Output

1. Classification analysis of the given specimen to its respective taxonomy rank:

   • Dataset
   • Family
   • Species

2. Image of the Dataset it belongs to.

3. DNA Barcode string converted to an image form where single character corresponds to a strip of colour code as:

   • A --> Green
   • C --> Blue
   • T --> Red
   • G --> Black

4. Performance Metrics of the chosen ML model for Label Prediction (Species and Family).

Technology

• Python
  • NumPy
  • pandas
  • scikit-learn
  • Matplotlib
  • Seaborn
  • fasta2csv : converting FASTA format to CSV
  • Tkinter : GUI
  • Pillow
  • os
  • time

Dataset

Collection

The Train and Test Datasets in FASTA format can be found in the data folder.

The data which we used was cleaned, merged and manipulated to match our needs. This was extracted from Empirical Datasets of research papers. Links for the Datasets:

• https://github.com/zhangab2008/BarcodingR/blob/master/Appendix_S2_empiricalDatasets.zip
• http://dmb.iasi.cnr.it/supbarcodes.php

Description

We are using datasets from 10 different organisms with multiple numbers of species within them. Table below gives the summary of the dataset.

No.	Dataset	#seq.	seq.length	#species	Gene region(s)
1	Bats	826	659	82	COI
2	Fish	626	419	82	COI
3	Birds	1936	703	575	COI
4	Amphibian	357	669	30	COI
5	Plants (Inga)	913	1,838	56	tmTD, ITS
6	Sea snail (Cypraeidae)	2,008	614	211	COI
7	Fruit Flies (Drosophila)	615	663	19	COI
8	Butterfly	1235	658	174	COI
9	Fungi	50	510	8	ITS
10	Algae	26	1,128	5	rbcL

Data after Cleaning

Raw FASTA files were cleaned by removing garbage values and redundant characters. They were then converted to CSV format for training the models.

Family	#sequences	#species
Algae	25	5
Amphibians	274	29
Bats	839	96
Birds	1396	574
Butterfly	926	174
Fish	625	82
Fruit Flies	498	19
Fungi	48	8
Plants	785	61
Sea Snail	1655	211

#sequences	#species

Performance Metrics

The 4 algorithms : Naïve Bayes, SVM, Random Forest and kNN are tested over the 4 performance metrics : Accuracy, Precison, Recall and F1-score to predict the Species and Family.
Total time taken for each algorithm to predict both the labels is considered as the Runtime of that algorithm.

Following Tables and graphs below indicate the results and further evaluation is comprehended on it’s basis.

Species Classification

	Naïve Bayes	SVM	Random Forest	kNN
Accuracy	0.890	0.878	0.901	0.871
Precision	0.965	0.969	0.967	0.964
Recall	0.978	0.971	0.985	0.966
F1	0.966	0.964	0.970	0.959

Family Classification

	Naïve Bayes	SVM	Random Forest	kNN
Accuracy	1.000	1.000	1.000	0.978
Precision	1.000	1.000	1.000	0.985
Recall	1.000	1.000	1.000	0.978
F1	1.000	1.000	1.000	0.980

Species Classification	Family Classification

Runtime

The specifications of the machine on which the implementations were performed and measured are:

• Model - MacBook Pro
• Processor - Intel R Core i5
• Installed RAM - 8 GB
• System type - 64-bit OS, x64 based processor
• OS - Windows 10 Pro

	Naïve Bayes	SVM	Random Forest	kNN
Runtime (in seconds)	9.288	202.471	86.966	3.609

Thus, Naïve Bayes seems to be the winner amongst the 4 algorithms for it's nearly best Perfromance metrics along with second best Runtime.

Screenshots

Setup

1. Clone this repository or download zip.

2. Open this repository on terminal. Navigate to src folder by typing cd src.

3. Type (if mentioned above python modules are not installed)

   pip install pandas sklearn matplotlib seaborn
   

4. To run the project,

    python main.py
   

5. All set.