Certificate Course on AI in Drug Development
Conducted by:
SOPHAS in Collaboration with Pumas-AI Inc
Duration:
September 2025 - April 2026
Teaching Methodology:
Flipped Classroom approach
Target Audience:
Background in statistics, evidence by academic coursework or certification in statistical methods relevant to pharmaceutical sector.
Objectives
- Understand the fundamental concepts of machine learning and their applications in drug development use cases.
- Understand and use machine learning techniques for covariate modelling and prognostic factor identification.
- Understand and use explainable machine learning techniques to answer drug development questions.
- Combine scientific knowledge and neural networks to build scientific machine learning (SciML) pharmacodynamic models.
- Understand the conditional variational autoencoder generative machine learning model and its relationship to nonlinear mixed effects models.
- Compare pure machine learning, traditional scientific modelling and hybrid SciML approaches when analyzing longitudinal clinical trial data with and without random effects.
- Build and use DeepNLME models to analyze disease progression and biomarker dataacross multiple realistic case studies.
Course Structure
Phase 1:
Access to course materials, lectures, and software for self-study and practice.
Duration: September 2025 – April 2026
Fees:
- Students – ₹2,00,000
- Academics – ₹2,50,000
- Industry – ₹4,00,000
Phase 2:
In-Person, 10-day workshop at a venue in South India.
Focus on Pumas software, including DeepPumas.
All topics from Phase 1 will be reinforced with a special focus on applications for drug
development.
Hands-on projects and case studies.
Duration: February 2nd Week, 2026
Phase 3:
Submission of assignments and participation in assessments.
Course completion certificates awarded upon successful completion.
Duration: April 2nd Week, 2026
Topics to be Covered:
Supervised Learning
- kNN (k-Nearest Neighbors)
- Random Forest
- Decision Tree
Unsupervised Learning
- kMedoids
ML Concepts
- Hyperparameters
- Cross-validation
- Confusion Matrix
- One-hot encoding
Explainability
- LIME (Local Interpretable Model-agnostic Explanations)
- SHAPELY (SHapley Additive exPlanations)
Neural Network Essentials
- Perceptron
- Activation functions
- MLP (Multi-Layer Perceptron)
Regularization
- LASSO (Least Absolute Shrinkage and Selection Operator)
NN in Low Data Regime
- Transfer Learning
- Fine-tuning
- Pre-training and self-supervision
- Data synthesis and augmentation
Time Series Modeling
- Recurrent NNs (RNN, LSTM, GRU)
- NeuralODE (Neural Ordinary Differential Equations)
- Scientific ODE Models
- SciML Neural ODE
Conditional Generative Models
- Conditional VAE (Variational Autoencoder)
- Scientific NLME (Non-Linear Mixed Effects modeling)
- DeepNLME
Instructors
Mohamed Tarek
PhD
