AI-Powered Insights in Pharmaceutical Sciences: Enhancing Drug Discovery and Development
Design by: Saptaparna Kuri, Aditi Shorma & Sourav Kar
Mentor: Radhika B Pandya, Assistant Professor (LMCP)
L.M. College of Pharmacy, Ahmedabad
Gujarat Technological University, Ahmedabad
Introduction Methodology Results & Conclusion
Artificial Intelligence (AI) is revolutionizing pharmaceutical sciences Databases: Model Results
by transforming how drugs are discovered, developed, and delivered. ChEMBL and PubChem (open-source)Input Format: SMILES
Traditional pharmaceutical R&D is often constrained by lengthy (Simplified Molecular Input Line Entry System) strings
•Accuracy: 91%
timelines, high costs, and high failure rates in clinical trials. AI Data Size: ~50,000 molecules used for training and validation • AUC-ROC Score: 0.89
introduces a paradigm shift by leveraging algorithms to identify drug Preprocessing: Canonicalization of SMILES • Precision: 0.87
candidates, predict pharmacokinetics and toxicity, and optimize Feature extraction using RDKit • Recall: 0.85
molecular interactions early in the pipeline. Conversion to molecular descriptors • Output Classes: Drug-like /
Through machine learning (ML) and deep learning models, vast Model Architecture:
amounts of biomedical and chemical data can be analyzed to Toxic/Inactive
Frameworks Used: Python, TensorFlow, RDKit, Scikit-learn
generate actionable insights. AI tools—such as convolutional neural Model Type: Convolutional Neural Network (CNN) + Multilayer
networks (CNNs), recurrent neural networks (RNNs), and generative Perceptron (MLP)
models—are employed to analyze compound structures, screen Training Split: 80% training, 20% testing
vast libraries, and simulate drug-target interactions. Optimization: Adam optimizer with cross-entropy loss
Moreover, AI enables virtual screening, automates lead optimization, Evaluation Metrics: Accuracy, ROC AUC Score, Precision, Recall
and supports personalized medicine by predicting patient-specific
responses to drugs. With platforms integrating cheminformatics, AI,
and bioinformatics, researchers can now identify viable compounds
in a fraction of the time and cost.
This project aims to showcase a deep learning-based approach that
integrates chemical datasets (e.g., PubChem, ChEMBL) with open- Interactive Streamlit Dashboard
source AI tools to improve drug discovery. The proposed solution
holds immense promise in reducing attrition rates, ensuring safer
clinical trials, and accelerating the journey from lab to patient.
Problem Statement
Lengthy and Expensive R&D: Drug development can take 10–15
years and cost over $2 billion.
High Clinical Failure Rates: Over 90% of drug candidates fail in
clinical trials.
Manual Screening Bottlenecks: Traditional molecule screening
is slow and resource-intensive.
Lack of Predictive Modeling: Early prediction of toxicity and Conclusion & Future Scope
bioactivity remains a major hurdle. Conclusion: This AI-powered model improves accuracy in
early drug screening, reduces R&D costs, and enhances
decision-making. It demonstrates a scalable, low-cost tool for
Objectives pharmaceutical research.
Future Scope:
Develop a deep learning model for predicting drug-likeness and • Expand model to multi-target prediction,
toxicity. •Integrate Explainable AI (XAI) for transparency
Integrate SMILES-based descriptors using tools like RDKit and Workflow: •Use Federated Learning for data privacy,
TensorFlow. Molecules in SMILES format are processed using RDKit to extract •Improve dashboard with visualizations
structural descriptors. These features are fed into a deep learning model
Significantly reduce time and cost associated with early drug
(CNN + MLP) to predict drug-likeness, toxicity, and bioactivity. Results are
References:
discovery.
displayed through an interactive Streamlit dashboard for real-time •Chen, H., Engkvist, O., Wang, Y., Olivecrona, M., Blaschke, T., “The Rise of Deep
Design a Streamlit dashboard for real-time AI prediction of Learning in Drug Discovery,” Drug Discovery
molecular behavior. insights.
Today, pp. 1241–1250, October 2018.
Demonstrate a scalable AI framework for pharmaceutical R&D •Paul, D., Sanap, G., Shenoy, S., Kalyane, D., Kalia, K., “Artificial Intelligence in
teams. Drug Discovery and Development,” Drug
Discovery Today, pp. 199–210, March 2021
