• image06


DrugBug predicts the biotransformation of drugs in human gut through microbial enzymes present in the gut microbes.


Human gut microbiota is constituted by a diverse group of microbial species having an enormous metabolic potential to alter the pharmacokinetic and pharmacodynamic properties of orally administered drugs. The limited knowledge of microbial metabolic activities in human gut has been a hurdle in estimating the biotransformation of drugs which is required to understand the observed individual/population-specific differences in drug responses. Thus, we have developed a robust and efficient high throughput method to predict the biotransformation of drugs in human gut through microbial enzymes.

Since the metabolic enzymes show promiscuity and thus are capable of metabolizing structurally similar substrates molecules, DrugBug exploits the molecular properties of known substrates of all metabolic enzymes (324,697) encoded by about 491 gut bacteria. Molecular information of all the substrates was translated into bit language (Fingerprints) and was utilized for preparing random forest (RF) classification models. Thus it employs both machine learning and chemoinformatic approaches for the prediction. DrugBug shows upto 94% sensitivity and >99% specificity on the training set and 75% sensitivity and 98% specificity on the blind set. DrugBug predicts the potential gut microbes and corresponding enzymes (EC number, metabolic reaction) responsible for the bio-transformation of the drugs. Considering the population, age, gender, etc., specific variations in the gut microflora, the knowledge of the gut microbe linked drug metabolism could help in predicting the case-specific metabolism of a drug which is significant for pharmacological studies and personalized medicine.

The flowchart of DrugBug approach is shown below.

The authors thank Mr. Vishnu Prasoodanan for making the website main figure.

Prediction       Steps

DrugBug predicts the microbial enzyme (EC, reaction), gut microbe (with taxonomy) for an input molecule using the following steps.
  1. Input molecule is first classified in any of the six EC classes using the RF model
  2. Assignment in EC sub-classes is made using EC class-specific RF model
  3. Four-digit EC number is assigned to input molecule using similarity search against the custom made EC sub-class database of the predicted EC sub-class
  4. The predicted EC number is used to identify the potential microbial enzyme and gut microbes using integrated Perl scripts