Tuberculosis (TB) remains to be a global major public-health threat, causing millions of deaths each year. A major difficulty in dealing with TB is that the causative bacterium, Mycobacterium tuberculosis, can persist in host tissue for a long period of time even after treatment. Mycobacterial persistence has become a central research focus for developing next-generation TB drugs. Latest genomic technology has enabled a high-throughput approach for identifying potential TB drug targets. Each gene product can be screened for its uniqueness to the TB metabolism, host-pathogen discrimination, essentiality for survival, and potential for chemical binding, among other properties. However, the exhaustive search for useful drug targets over the entire genome would not be productive as expected in practice. On the other hand, the problem can be formulated as pattern recognition or inductive learning and tackled with rule-based or statistically based learning algorithms. Here we review the perspective that combines machine learning and genomics for drug discovery in tuberculosis.