Machine Learning and Molecular HLA Mismatching to Improve Organ Allocation and Immunological Risk Stratification in Paediatric Kidney Transplantation

Kidney transplantation is the best treatment for patients whose kidneys have stopped working. Patients feel better as they have better kidney function, and they have more independent lives without needing to perform regular dialysis. However, transplant patients need to take medications which have many side effects including a lower immune system. These medications (called immunosuppression) are important to stop the body 'rejecting' the transplant. The immune system recognises the part of the kidney called HLA and attacks the transplant. We all have different HLA or tissue types. The different HLA types are divided into multiple groups. However, we can now identify the exact high resolution HLA type using modern gene sequencing technology. Our lab has developed computer algorithms which compare the molecular sequence and structure of HLA to determine how different (or similar) the transplant is to the patient. We showed that the molecular mismatching scores are good at predicting rejection and antibodies against the transplant. In this study, we want to apply the molecular mismatching to children and young adults up to 35 years of age because they have high rates of infection and rejection. In the first part, we will look at how molecular mismatching can be used in the kidney allocation system. We will use modern machine learning tools to predict which transplants are more likely to stop working early. We will compare current HLA mismatching against the new molecular mismatching. We will then perform a computer simulation using information from the donors and kidney recipients that were transplanted in the last ten years. We are working on a machine learning model which can better allocate transplants to the most appropriate patient, so that each transplant lasts as long as possible for that particular patient. We want to check if the molecular scores affect waiting times. We believe that it would be easier to find a better matched kidney this way because it allows comparisons between all HLA groups. In the second part, we want to see if molecular mismatching can be used to guide what treatment children get. It would beneficial if we could reduce the medication for children with good tissue matches. We have close links with the international CERTAIN registry which contains valuable information on the treatment that children receive over time and the outcomes of their transplants. We will use machine learning again to predict how different levels and doses of immunosuppression affect rejection over time. We are conscious that machine learning methods can contain hidden biases. We will form a diverse group of doctors and include patients in this group to oversee our research. We will specifically check the impact of the algorithms on different patient groups, for example patients from ethnic minority backgrounds. We want to produce machine learning tools which are fair and transparent so that they can be used safely in day-to-day practice. In conclusion, this study combines advanced molecular tools with machine learning to improve the chances of successful kidney transplants. Better tissue matching means less risk of rejection and fewer side effects from medication.