PhD Studentship: Equitable AI for Health Predicting Cardiometabolic Disease in Underserved Populations

Cardiometabolic diseases (CVMD), such as heart disease and type 2 diabetes, represent a major global health burden and exhibit stark ethnic disparities. Current clinical prediction models, even those using advanced AI, often fail to capture the nuanced clinical and social factors driving these differences, leading to inequitable health outcomes. This is largely because they struggle to leverage the rich, unstructured information found in clinical notes and cannot effectively gather data on lifestyle and social determinants of health.

This PhD project will pioneer a novel, hybrid AI framework to directly address this critical gap. The research aims to develop a more accurate and equitable tool for early CVMD risk prediction, with a specific focus on the underserved South Asian population, who carry a disproportionately high disease burden.

The project is built on a powerful two-part methodology:

1. Culturally-Adapted Data Acquisition: You will fine-tune a multilingual LLM to interact with patients in their native languages. This conversational AI will gather crucial, yet often unrecorded, data on symptoms, lifestyle, and social determinants of health, converting unstructured dialogue into standardized features.
2. Transformer-Based Risk Prediction: A central component of this project involves developing your own time-aware transformer model. This predictive back-end will be trained on world-leading, large-scale biomedical datasets, including UK Biobank and the CPRD. It will integrate complex longitudinal data—including harmonized EHRs, genetics, and the novel features captured by the LLM—to generate personalised, ethnically-stratified risk scores.

This is a highly interdisciplinary project at the intersection of machine learning, health equity, and precision medicine. The successful candidate will join a vibrant research environment and gain extensive, hands-on experience in:

• Cutting-Edge AI: Developing and implementing your own custom transformer models tailored for clinical data, alongside fine-tuning existing biomedical foundation models (e.g., Med-PaLM) using techniques like Low-Rank Adaptation (LoRA).
• Big Data Analytics: Managing and analysing complex, multi-modal data from globally significant resources like UK Biobank.
• Advanced Techniques: Applying causal inference frameworks, unsupervised learning for phenotype discovery, and population-specific Genome-Wide Association Studies (GWAS).
• Ethical & Explainable AI (XAI): Implementing techniques such as SHAP and LIME to ensure model transparency and conducting rigorous, ethnicity-stratified evaluations to mitigate bias.

We are seeking a highly motivated candidate with a strong quantitative background (e.g., in computer science, statistics, bioinformatics). The following skills are essential for this project:

• Excellent programming skills in Python.
• Proven experience with deep learning frameworks (e.g., PyTorch, TensorFlow).
• Direct, hands-on experience working with Large Language Models (LLMs) and/or transformer models.
• Familiarity and experience working with Electronic Health Record (EHR) data.

Knowledge of analysing genomic data is highly desirable albeit not essential. This project offers a unique opportunity to develop state-of-the-art AI solutions that can create a tangible impact by reducing health inequalities and improving clinical outcomes for all populations.

This is a PhD studentship with the Midlands Integrated Biosciences Training Partnership, funded by BBSRC.

References:

Rashid et al. ALPK1 hotspot mutation as a driver of human spiradenoma and spiradenocarcinoma.

Jackstadt et al. Epithelial NOTCH signaling rewires the tumor microenvironment of colorectal cancer to drive poor-prognosis subtypes and metastasis.

Lo JA et al. Epitope spreading toward wild-type melanocyte-lineage antigens rescues suboptimal immune checkpoint blockade responses.

Walker et al. Hydroxymethylation profile of cell-free DNA is a biomarker for early colorectal cancer.

Kabir et al. Automatic speech recognition for biomedical data in bengali language.

Advert information