Drug function identification from the drug
properties is important in drug discovery. Each year
billions of dollars are spent on empirical testing of the
drugs, which is costly, chemical wastage, and timeconsuming. The computational experiments would help
reduce drug discovery time and cost significantly. Most of
the existing works have focused on single-label drug
function identification. However, the capability of the
drug's biological properties (transporter, target, carrier,
and enzyme) has not yet been explored for multiple drug
function identification. Identifying drug function is a
multi-label classification problem. So, in the present work,
a multi-label long short-term memory-based
framework has been proposed for identifying drug
function. The data related to biological properties has been
extracted from DrugBank, and drug functions are
collected from PubChem. The proposed framework
performance has been found promising in terms of
accuracy, precision, recall, F1, ROC-AUC score, and
hamming-loss, and it achieved the highest accuracy of
95.80%.
Keywords :
Multi-Label, LSTM, Biological Properties, Drug Function, Machine Learning.