Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), is one of the oldest known and most dangerous diseases. Although the spread of TB was controlled in the early 20th century using antibiotics and vaccines, TB has again become a threat because of increased drug resistance. There is still a lack of effective treatment regimens for a person who is already infected with multidrug-resistant Mtb (MDR-Mtb) or extensively drug-resistant Mtb (XDRMtb). In the past decades, many research groups have explored the drug resistance profiles of Mtb based on sequence data by GWAS, which identified some mutations that were significantly linked with drug resistance, and attempted to explain the resistance mechanisms. However, they mainly focused on several significant mutations in drug targets (e.g. rpoB, katG). Some genes which are potentially associated with drug resistance may be overlooked by the GWAS analysis. In this article, our motivation is to detect potential drug resistance genes of Mtb using a heat diffusion model. All sequencing data, which contained 127 samples of Mtb, i.e. 34 ethambutol-, 65 isoniazid-, 53 rifampicin- and 45 streptomycin-resistant strains. The raw sequence data were preprocessed using Trimmomatic software and aligned to the Mtb H37Rv reference genome using Bowtie2. From the resulting alignments, SAMtools and VarScan were used to filter sequences and call SNPs. The GWAS was performed by the PLINK package to obtain the significant SNPs, which were mapped to genes. The P-values of genes calculated by GWAS were transferred into a heat vector. The heat vector and the Mtb protein-protein interactions (PPI) derived from the STRING database were inputted into the heat diffusion model to obtain significant subnetworks by HotNet2. Finally, the most significant (P < 0.05) subnetworks associated with different phenotypes were obtained. To verify the change of binding energy between the drug and target before and after mutation, the method of molecular dynamics simulation was performed using the AMBER software. We identified significant subnetworks in rifampicin-resistant samples. Excitingly, we found rpoB and rpoC, which are drug targets of rifampicin. From the protein structure of rpoB, the mutation location was extremely close to the drug binding site, with a distance of only 3.97 Å. Molecular dynamics simulation revealed that the binding energy of rpoB and rifampicin decreased after D435V mutation. To a large extent, this mutation can influence the affinity of drug-target binding. In addition, topA and pyrG were reported to be linked with drug resistance, and might be new TB drug targets. Other genes that have not yet been reported are worth further study. Using a heat diffusion model in combination with GWAS results and protein-protein interactions, the significantly mutated subnetworks in rifampicin-resistant samples were found. The subnetwork not only contained the known targets of rifampicin (rpoB, rpoC), but also included topA and pyrG, which are potentially associated with drug resistance. Together, these results offer deeper insights into drug resistance of Mtb, and provides potential drug targets for finding new antituberculosis drugs.