AI-Assisted Predictive Toxicology: QSAR, ADMET, hERG & Organ-Toxicity Screening for Drug Safety
From Molecular Structure to Safety Prediction — Use AI and QSAR to Predict Drug Toxicity, ADMET Risks and Organ-Specific Safety Profiles.
Skills you will gain:
About Program:
This 3-day mentor-led workshop introduces participants to AI-assisted predictive toxicology for early-stage drug safety assessment. Participants will learn how molecular structure and chemical descriptors can be transformed into predictive models for toxicity, ADMET properties, hERG liability, and organ-specific risks. The workshop combines QSAR, cheminformatics, machine learning, molecular descriptors, model interpretation, and virtual screening to demonstrate how computational approaches can support safer compound prioritisation before experimental validation.
Aim: To provide practical, research-oriented training in AI-assisted predictive toxicology, enabling participants to build and interpret computational workflows for QSAR, ADMET, hERG liability, and organ-toxicity screening in drug discovery.
Program Objectives:
- Understand the principles of predictive toxicology and QSAR in modern drug discovery.
- Generate and curate molecular descriptors and chemical fingerprints for toxicity modelling.
- Develop machine-learning models to predict toxicity and key ADMET properties.
- Assess potential hERG-related cardiac liability using computational approaches.
- Screen compounds for potential hepatotoxicity, nephrotoxicity, neurotoxicity, and general toxicity.
- Apply explainable AI to identify molecular features contributing to predicted toxicity.
- Integrate QSAR, ADMET, hERG, and organ-toxicity predictions into a unified safety profile.
- Prioritise drug candidates based on predicted safety and risk profiles.
- Interpret model performance, applicability domains, uncertainty, and limitations.
- Build a reproducible AI-assisted workflow for early-stage computational drug-safety assessment.
What you will learn?
📅 Day 1: QSAR, Molecular Descriptors & AI-Based Toxicity Prediction
Focus: Transform chemical structures into machine-learning-ready datasets and develop predictive QSAR models for toxicity assessment.
Topics Covered
- Introduction to predictive toxicology and its role in drug discovery
- Structure–toxicity relationships and QSAR principles
- SMILES, molecular structures, fingerprints, and molecular descriptors
- Physicochemical and structural descriptors for toxicity modelling
- Dataset curation, endpoint definition, and data-quality considerations
- Handling missing data, duplicates, outliers, and class imbalance
- Feature selection and dimensionality reduction
- Classification and regression approaches for toxicity prediction
- Random Forest, XGBoost and other ML approaches
- Model evaluation using accuracy, precision, recall, F1-score, ROC-AUC, MAE and RMSE
- Applicability domain and limitations of predictive models
🛠️ Hands-on
Build a QSAR toxicity dataset from chemical structures, generate molecular descriptors/fingerprints, train an ML model, evaluate its performance, and predict toxicity for new compounds.
Workflow:
Chemical Structures → Curation → Molecular Descriptors → Feature Selection → QSAR/ML Model → Validation → Toxicity Prediction
Tools: RDKit, Python, pandas, NumPy, scikit-learn, XGBoost, Google Colab
📅 Day 2: ADMET & hERG Liability Prediction
Focus: Predict drug-like properties and major safety liabilities relevant to pharmacokinetics and cardiac safety.
Topics Covered
- ADMET fundamentals and their importance in drug development
- Absorption, distribution, metabolism, excretion and toxicity
- Solubility, permeability, lipophilicity and related molecular properties
- Blood–brain barrier and distribution-related prediction
- Metabolic stability and CYP-related considerations
- Drug-drug interaction and transporter-related safety concepts
- Plasma protein binding and clearance concepts
- Introduction to hERG channel liability and cardiac safety
- hERG inhibition prediction using QSAR and machine learning
- Classification of compounds according to predicted safety risk
- Comparing predicted ADMET and toxicity profiles across compounds
- Interpreting model predictions and uncertainty
🛠️ Hands-on
Generate molecular descriptors, build an ADMET/hERG prediction workflow, classify compounds according to predicted risk, and compare safety profiles.
Workflow:
Molecular Structures → Descriptors → ADMET Prediction → hERG Model → Risk Classification → Safety Profile
Tools: RDKit, PubChem, ChEMBL, Python, scikit-learn, XGBoost, Google Colab
📅 Day 3: Organ-Toxicity Screening, Explainable AI & Compound Prioritisation
Focus: Integrate multiple toxicity endpoints to identify potential organ-specific risks and prioritise compounds for further investigation.
Topics Covered
- Organ-specific toxicity and major safety endpoints
- Hepatotoxicity and liver-injury prediction
- Nephrotoxicity and kidney-related safety assessment
- Neurotoxicity and CNS-related toxicity concepts
- General cytotoxicity and systemic toxicity
- Multi-endpoint toxicity profiling
- Integrating QSAR and ADMET predictions for compound assessment
- Explainable AI for toxicity prediction
- SHAP-based feature importance and interpretation
- Identifying molecular features associated with toxicity
- Multi-parameter safety-risk assessment
- Virtual toxicity screening and compound prioritisation
- Distinguishing computational predictions from experimentally validated toxicity
- Building an interpretable computational safety report
🛠️ Hands-on
Generate organ-toxicity predictions, interpret important molecular features using explainable AI, integrate multiple safety endpoints, and rank compounds according to their predicted risk profile.
Workflow:
Compound Library → Toxicity Prediction → Organ-Specific Screening → SHAP/XAI → Multi-Endpoint Integration → Risk Profiling → Compound Prioritisation
Tools: RDKit, Python, scikit-learn, XGBoost, SHAP, pandas, Plotly, Google Colab
Mentor Profile
Fee Plan
Get an e-Certificate of Participation!
Intended For :
- Graduate and postgraduate students in Biotechnology, Biochemistry, Chemistry, Pharmacy, Life Sciences, Bioinformatics, and related disciplines.
- PhD scholars and research scholars working in drug discovery, medicinal chemistry, computational toxicology, pharmacology, ADMET, or cheminformatics.
- Academicians and faculty members interested in AI-assisted drug safety and predictive toxicology.
- Industry professionals from pharmaceutical, biotechnology, CRO, drug discovery, toxicology, and computational research sectors.
- Professionals and researchers interested in QSAR, machine learning, ADMET prediction, hERG screening, and organ-toxicity assessment.
Career Supporting Skills
Program Outcomes
- Explain the principles of predictive toxicology and QSAR.
- Convert chemical structures into computationally useful molecular representations.
- Generate molecular descriptors and fingerprints using cheminformatics tools.
- Curate and preprocess toxicity datasets for machine learning.
- Build and evaluate QSAR-based toxicity prediction models.
- Interpret classification and regression performance metrics.
- Predict key ADMET properties computationally.
- Understand and model hERG-related cardiac safety liability.
- Screen compounds for potential liver, kidney, CNS and general toxicity risks.
- Apply explainable AI to understand toxicity predictions.
- Identify molecular features associated with predicted toxicity.
- Integrate multiple safety endpoints into a compound risk profile.
- Prioritise compounds for further experimental investigation.
- Understand the limitations, uncertainty and applicability domain of computational toxicity models.