Implementation of SMILES2Vec-based LSTM for Predicting Drug Side Effects: Case Study of Hepatobiliary Disorder
Ananda Fitri Karimah, Angel Metanosa Afinda, Isman Kurniawan · 2023
Drug is a molecule that interacts with a specific target protein to intervene biological systems through numerous molecular interactions. However, drug molecules can interact with non-target proteins, leading to the possibility of side effects. Hence, it is necessary to be able to predict the side effects to prevent more severe conditions. Several approaches are commonly used to predict side effects. However, these approaches have several weaknesses, such as safety procedure requirements that are pretty minimal and less accurate. In-silico approach that uses numerical and computational tools can be implemented as an alternative method in predicting side effects. This approach has advantages in reducing experimental process risks. This study aims to implement SMILES based descriptor, or commonly called SMILES2Vec, of drug molecule on LSTM method to predict side effect, for the case study of hepatobiliary disorder. SMILES based representation of descriptor is preferable because the calculation is faster than other descriptor, such as 2D, 3D, fingerprint, etc. Meanwhile, Long Short-Term Memory (LSTM) was used for the prediction since this method is suitable with 1D type of SMILES based descriptor. The architecture of LSTM was determined by using manual tuning approach by determining several schemes. According to the result, we found that LSTM architecture with a combination of CONV and 2-layer of LSTM using Adagrad or Adadelta optimizer gives the best performance with values of F-1 score 68.62% for both.