File	Drug	Test AUC-ROC	CV AUC-ROC	Size
rf_RIFAMPICIN_v2.joblib	Rifampicin	0.975	0.969 ± 0.004	61MB
rf_ISONIAZID_v2.joblib	Isoniazid	0.948	0.946 ± 0.008	48MB
rf_ETHAMBUTOL_v2.joblib	Ethambutol	0.894	0.900 ± 0.007	77MB
rf_PYRAZINAMIDE_v2.joblib	Pyrazinamide	0.886	0.883 ± 0.007	58MB

Usage

python

from huggingface_hub import hf_hub_download
import joblib

# Download and load a model
path = hf_hub_download(
    repo_id="nanzhen102/FORUM-TB-models",
    filename="rf_RIFAMPICIN_v2.joblib"
)
rf = joblib.load(path)

Input Format

Models expect a feature vector of 2,693 AMR gene positions encoded as integers (0=REF, 1=A, 2=T, 3=C, 4=G).

Download the ML-ready dataset directly from Kaggle: https://www.kaggle.com/datasets/nanzhen/forum-tb?resource=download

Biological Validation

Top SHAP features confirmed against known resistance mutations:

pos_761155 → rpoB codon 450 → S450L ✅ (Rifampicin)
pos_2155168 → katG codon 315 → S315T ✅ (Isoniazid)
pos_4247429 → embB codon 306 → M306I/V ✅ (Ethambutol)

Authors

Noah LeGall, Ph.D. — The Microbialist
Nanzhen (Aspen) Qiao — Queen's University, Kingston, Canada

License

CC BY-NC 4.0 — free for non-commercial use with attribution.