What is BDQR-MTB?

BDQR-MTB is a freely available web-based tool for machine learning based prediction of BDQ resistance in MTB clinical isolates. The prediction can be done for a single sample VCF file, which will be given as input.

Dataset

• Whole genome sequencing (WGS) data of BDQ-resistant and susceptible MTB clinical isolates.
• Data from 16 different Bioproject IDs, comprising 632 samples/run IDs. The dataset includes BDQ-resistant samples (n=294) and susceptible samples (n=340).

Features

• BDQR-MTB consists of:
• The complete optimal model consisting of a total of 8282 features.
• The minimalistic optimal model trained with 50 features.
• The SHAP explainer model.
• The input page allows users to upload VCF files of MTB clinical isolates.
• The output page of BDQR-MTB gives:
• The prediction result for the uploaded VCF file.
• The prediction probabilities based on the full optimal model.
• The top 20 contributing features as obtained from the application of SHAP on the minimalistic optimal model.

Data Availability

• The details of the BioProject IDs are available here.
• The run IDs considered in this study are available here.