Long Short Term Memory model for accurate predictions of Single Nucleotide Polymorphisms in Mycobacterium tuberculosis from timeseries genome analysis

Drug resistance in tuberculosis is on the serious threat list of world health organization with a critical focus on addressing the genomic variations in M ycobacterium tuberculosis. This provides an opportunity for better understanding of evolutionary progression leading to anti-microbial resistance. This impacts on the economic stability of the global healthcare sector. A timeline genomic analysis from 2003 to 2021 of 578 mycobacterium genomes have been performed to understand the pattern underlying the genomic variations. A total of 4,76,053 mutations with Ts/Tv ratio of 0.448 was observed. In this regard, a recurrent neural network approach of Long short term memory model was optimized to predict the genome-wide mutations in the ratio of 80:20 test to training set respectively. The genomic sequences were split into batches and the error rate is averaged to 5.14% in 4 out of 5 batches and 38.99% in last batch with accurate position specific predictions. This has an impact on countering the anti-microbial resistance by identifying regions with high and low genomic variability providing insights on novel drug prospective targets. The further scope lies in improvising the model by enriching the datasets.