ABSTRACT
The HIV (Human Immunodeficiency Virus) drug resistance database of Stanford University provides the resistance exhibited by HIV towards drugs administered based on its mutation patterns. Numerous deep learning and machine learning models have been considered to perform drug prediction in previous studies. These models are expected to establish associations and infer patterns from the training dataset. The inferred information should be used to perform informed predictions on the test combinations of mutants, which the models have never witnessed. Given that the mutants’ major biological mutation positions are limited, apart from prediction, it is essential to understand and make inferences on the behavior of individual mutants and mutant combinations concerning drugs. These quantifiable relationships could have remarkable significance for medical researchers. The proposed method in this work derives probabilities from the dataset, which gives insights into the chances of expecting a mutant given a drug administered. These conditional probabilities have established a way to predict the drug in the presence of a mutation. The combination of conditional and joint probabilities serves the purpose of establishing better probabilities on the most suitable drugs for mutation patterns observed in a patient to perform a recommendation. The difference in resistances, the key source to compare the performance of drugs on mutation patterns, is well accepted to consider the proposed method of statistical analysis.
