Colorectal Cancer Prediction with Gut Microbiome Data

This project builds a machine learning model to predict colorectal cancer (CRC) based on the gut microbiome composition of patients, using 16S rRNA sequencing data collapsed at the genus level.

---

Project Workflow

1. Load the Data

• metadata.csv: Includes SampleID and whether the patient has CRC.
• seqtab_nochim.csv: Rows = samples, Columns = ASVs (amplicon sequence variants), Values = counts.
• taxa.csv: Maps ASVs to their taxonomic classification (e.g., Genus, Family).

---

2. Map ASVs to Genera

• Use the taxa.csv table to map each ASV to its genus.
• Missing genera are labeled as Unassigned.
• Rename ASV columns from IDs (e.g., ASV_1234) to genus names (e.g., Bacteroides).

---

3. Collapse to Genus Level

• Some ASVs belong to the same genus.
• Combine all columns with the same genus name by summing their counts per sample.

---

4. Join with Metadata

• Merge CRC status (from metadata.csv) with genus-level abundances.

• Save as seqtab_genus.

• Final table structure:

  • Rows = samples (patients’ stool samples).
  • Columns = bacterial genera (e.g., Bacteroides, Prevotella).
  • Values = read counts per genus.

---

5. Normalize the Microbiome Data

• Convert raw counts to relative abundances (percentage per sample).

• Prepare:

  • X = features (genus abundances)
  • y = labels (CRC status)

---

6. Split the Dataset

• Perform an 80/20 train-test split.

---

7. Train a Random Forest Classifier

• Fit a Random Forest model to learn relationships between microbial genera and CRC status.
• Uses an ensemble of decision trees.

---

8. Evaluate Model Performance

• Report:

  • Precision
  • Recall
  • F1-score
  • Accuracy

---

9. Identify Important Genera

• Extract feature importance from the Random Forest.
• Highlight genera most predictive of CRC status.

---

Dependencies

• Python 3.8+
• pandas, numpy, scikit-learn, matplotlib/seaborn (for analysis & visualization)

---

Usage

1. Place the input files (metadata.csv, seqtab_nochim.csv, taxa.csv) in the working directory.
2. Run the preprocessing script to generate seqtab_genus.
3. Train the Random Forest model.
4. Evaluate results and inspect important genera.

---

Output

• Trained model performance metrics.
• List of key microbial genera predictive of CRC.

---

License

This project is open-source and available under the MIT License. See the LICENSE file for details.