Taxon-Classifier

A FASTQ-based machine learning classifier that identifies Anopheles mosquito species directly from raw sequencing reads — no alignment or variant calling required.

Built as part of a MalariaGEN GSoC 2026 proposal for lightweight taxonomic identification of malaria vectors.

What is it?

Taxonomic identification of Anopheles mosquitoes is critical for malaria control but challenging because many species are morphologically identical. Misidentification leads to wasted sequencing resources and flawed epidemiological insights.

This tool classifies mosquito samples into three major species groups using only raw FASTQ reads:

Species	Region
An. gambiae complex	Sub-Saharan Africa
An. funestus	Sub-Saharan Africa
An. stephensi	South/Southeast Asia

Features

• No alignment needed — works directly on raw FASTQ reads streamed from ENA
• K-mer frequency profiles — uses HashingVectorizer to extract character n-gram features
• LightGBM classifier — fast, handles sparse high-dimensional data natively
• Multiple k-mer sizes — tested with k=15 (98.3% accuracy) and k=21 (97.5% accuracy)
• Noise robust — maintains 100% accuracy up to 10% base substitution error rate
• Low read depth — reliable classification with as few as 1,000 reads

Results

Metric	k=15	k=21
Accuracy	98.3%	97.5%
Macro-F1	0.983	0.975
Compression Ratio	10³:1	10⁶:1

How it works

Raw FASTQ reads → K-mer extraction → Hashing (2²⁰ bins) → LightGBM → Species prediction

1. Extract: Stream FASTQ reads from ENA, extract k-mer frequency profiles using HashingVectorizer
2. Train: Train a LightGBM model with stratified cross-validation
3. Evaluate: Predict species from unseen FASTQ files with confidence scores

Local Setup

1. Clone the repository

git clone https://github.com/31puneet/Taxon-Classifier.git
cd Taxon-Classifier

2. Create and activate virtual environment

python -m venv venv
source venv/bin/activate        # Linux/Mac
# venv\Scripts\activate         # Windows

3. Install dependencies

pip install -r requirements.txt

4. Run evaluation (pre-trained models included)

cd src/evaluation
python evaluate_models.py --k 15

5. Run noise injection test

python poison_test.py --k 15

6. Run the full pipeline (requires internet for ENA streaming)

cd src
python main.py --k 15

Or run individual steps:

cd src/scripts
python extract_kmers.py --k 15
python train_models.py --k 15
cd ../evaluation
python evaluate_models.py --k 15

Visualizations

Open notebooks/visualization.ipynb in Jupyter to generate plots:

from evaluation.visualize import plot_tsne, plot_k_comparison, plot_poison_comparison, plot_depth_curve

plot_tsne(k=15)               # t-SNE species clustering
plot_k_comparison()           # k=15 vs k=21 comparison
plot_poison_comparison()      # Noise robustness curves
plot_depth_curve(k=15)        # Read depth vs confidence