BP301T · Core · Semester 3
Introduction to Machine Learning in Pharmaceutical Sciences
Explore AI, machine-learning, and data-science concepts through transparent fictional pharmaceutical teaching cases, from data framing and evaluation to regression, classification, trees, ensembles, and clustering.
Course objectives
- Introduce fundamental Artificial Intelligence, Machine Learning, and Data Science concepts relevant to pharmaceutical sciences.
- Develop conceptual understanding of model building, training–testing strategies, and evaluation.
- Explain regression-model principles for continuous pharmaceutical and healthcare teaching outcomes.
- Familiarize learners with classification and tree-based models used in clearly simulated clinical-prediction contexts.
- Introduce unsupervised learning for transparent pattern-recognition and healthcare-data analysis examples.
Unit 1: Foundations of Machine Learning
Artificial Intelligence, Machine Learning, and Data Science use different scopes. A foundational ML workflow frames features and labels, separates training and testing data, and considers overfitting, underfitting, bias–variance trade-off, and the learning type before any result is interpreted.
Map a fictional formulation-learning problem into features, a label, a learning type, and a train/test split; then state one overfitting safeguard.
Unit 2: Regression Models in Healthcare
Linear and multiple linear regression describe continuous teaching outcomes. Inspect coefficients, residuals, MSE, RMSE, and R² before reading a fictional dose-response, dissolution-rate, or PK-style relationship; a fitted model does not establish clinical validity.
Fit an educational dose-response line, compare it with the best fit, inspect residuals and error metrics, and state why the result is not a real prediction.
Unit 3: Classification Models in Clinical Applications
Logistic regression, probability outputs, threshold choice, a confusion matrix, accuracy, sensitivity, specificity, precision, recall, ROC/AUC, and k-nearest-neighbor concepts help learners reason about categorical teaching outputs. No browser result is a clinical classifier.
Change a fictional screening threshold and explain the sensitivity–specificity trade-off, confusion-matrix entries, and no-clinical-use boundary.
Unit 4: Tree-Based Models & Ensemble Learning
Decision trees use splitting criteria, decision paths, and feature-importance concepts. Random forests introduce an ensemble idea; they may stabilize a teaching pattern while making explanation more complex. Advantages and limitations must be stated before applying a case context.
Describe a hypothetical first split for a fictional ADR-risk, patient-classification, or treatment-outcome teaching example and name an interpretability limitation.
Unit 5: Unsupervised Learning & Practical Case Studies
Unsupervised learning finds structure without target labels. K-means clustering and cluster-count selection can support a fictional patient-segmentation or drug-property grouping lesson, but a cluster is not a diagnosis, a treatment recommendation, or a validated clinical subgroup.
Choose a cluster count for a simulated segmentation or drug-grouping plot and explain what the clusters can and cannot represent.