Enhancing Drug Safety with Explainable Artificial Intelligence and Multimodal Prediction Models (RDF25/EE/CIS/JAGANATHAN)

Developing a new drug in the UK typically takes 10 to 15 years and costs around £1.15 billion. Despite the investment, about 90% of drug candidates fail in clinical trials. The main reason, accounting for about 30%, is drug-induced toxicity, particularly to important organs such as heart (cardiotoxicity), lungs (respiratory toxicity), liver (hepatotoxicity), and kidneys (nephrotoxicity). Doxorubicin is linked to cardiotoxicity, and paracetamol to liver toxicity. These toxic effects not only pose significant health risks but also contribute to the high withdrawal rates in drug development. Toxicity issues are often identified late in the development process, during clinical trials or after the drug has already reached the market, resulting in post-market withdrawal and increased costs.

This project addresses these challenges through the application of Explainable Artificial Intelligence (XAI), which is not only critical for improving early-stage toxicity prediction accuracy, but also integral to providing insights into the specific molecular features driving these predictions. XAI techniques are embedded to reveal the molecular features that contribute to toxicity predictions. This ensures accurate and interpretable predictions, enabling informed decisions. Traditional models rely on single data modalities, like chemical structures, failing to capture complex molecular interactions. To overcome this problem, this proposal introduces the Multimodal Molecular Fusion Model, an innovative AI framework that combines features from distinct molecular representations such as molecular graphs, numerical descriptors, fingerprints, molecular images and structural sequences. This approach will provide more accurate and early predictions. By integrating advanced deep learning techniques, such as Graph Neural Networks (GNN) and Transformer models, with SHAP and LIME explainability methods, this project ensures the predictions are not only accurate but also interpretable for clinicians.

The project objectives are:

Objective 1: Data Collection from Public Databases and Literature

a) Collect molecular and toxicity data from databases such as PubChem, ChEMBL, and LiverTox, focusing on drug-induced toxicities.

b) Gather diverse molecular data, including structures, descriptors, and SMILES sequences, for model training.

c) Ensure the quality and relevance of the collected data for accurate drug safety evaluations.

Objective 2: Development of a Multimodal AI Model for Drug Toxicity Prediction

a) Design and develop an AI model that integrates various molecular data types to predict toxic effects across organs (heart, liver, kidneys, lungs).

b) Train the model on diverse datasets to assess its accuracy.

c) Evaluate how the integration of different molecular representations improves model performance.

Objective 3: Model Validation, Generalization, and Explainability

a) Validate the model using independent datasets to ensure its generalizability across different toxicity profiles.

b) Measure performance using key metrics such as accuracy and precision to ensure the model’s reliability in real-world applications.

c) Utilize Explainable AI techniques, such as SHAP and LIME,  to make the model’s predictions transparent and interpretable. These techniques will provide insights into the molecular features driving toxicity predictions, ensuring clinicians and researchers make well-informed decisions. By integrating XAI, the project bridges the gap between prediction accuracy and real-world applicability in drug safety assessments.

This project lies at the intersection of AI, drug safety, and ethical research. By applying advanced AI techniques and adhering to the 3Rs principle (Replace, Reduce, Refine), it promotes a more sustainable and ethical approach to toxicity assessment, reducing reliance on animal testing. XAI is a core component of this project, playing an essential role in bridging the gap between predictive accuracy and interpretability. By providing transparency,  XAI fosters trust in predictions, enabling real-world applications in drug safety assessments. This project aims to reduce drug development time and costs while improving drug safety. This research will generate valuable insights for academic and industry, driving innovation in healthcare while promoting sustainable practices in drug discovery.

Academic Enquiries

This project is supervised by Dr Keerthana Jaganathan. For informal academic queries, contact keerthana.jaganathan@northumbria.ac.uk . For enquiries relating to eligibility or application process, email fc.pgr.admissions@northumbria.ac.uk

Eligibility Requirements:

•       Academic excellence i.e. 2:1 (or equivalent GPA from non-UK universities with preference for 1st class honours); or a Masters (preference for Merit or above); or APEL evidence of substantial practitioner achievement.

•       Appropriate IELTS score, if required.

•       Applicants cannot apply if they are already a PhD holder or if currently engaged in Doctoral study at Northumbria or elsewhere.

•       Must be able to commit to campus-based full-time or part-time study.

To be classed as a Home student, candidates must:

•       Be a UK National (meeting residency requirements), or

•       have settled status, or

•       have pre-settled status (meeting residency requirements), or

•       have indefinite leave to remain or enter.

If a candidate does not meet the criteria above, they would be classed as an International student and not eligible to apply for this studentship. Applicants will need to be fully enrolled in the UK before stipend payments can commence.

Apply now at https://www.northumbria.ac.uk/research/postgraduate-research-degrees/how-to-apply/  

In your application, include a research proposal of approximately 1,000 words and also the advert reference (e.g. RDF25/…).

Application Deadline: 24 January 2025

Start date of course: 1 October 2025

Northumbria University is committed to creating an inclusive culture where we take pride in, and value, the diversity of our postgraduate research students. We encourage and welcome applications from all members of the community. The University holds a bronze Athena Swan award in recognition of our commitment to advancing gender equality, we are a Disability Confident Leader, a member of the Race Equality Charter and are participating in the Stonewall Diversity Champion Programme. We also hold the HR Excellence in Research award for implementing the concordat supporting the career Development of Researchers and are members of the Euraxess initiative to deliver information and support to professional researchers.