Machine Learning as a Tool to Predict Early and Chronic Graft Dysfunction after Lung Transplantation
PI:
Jason Cory Brunson, PhD
Project Summary
Lung transplantation is the treatment of last resort for a large number of patients with end-stage lung diseases, but is complicated by a modest post-transplant survival. Mortality after transplantation is mostly driven by chronic lung allograft dysfunction (CLAD). An early complication of lung transplantation, primary graft dysfunction (PGD), is a major risk factor for subsequent development of CLAD. Tools to more accurately predict these complications are urgently needed. This proposal hypothesizes that integrating clinical and biological markers from donors and recipients will predict the incidence and severity of PGD and CLAD. This award will support the career development of a mathematician–data scientist toward a critical and underexplored area. The candidate has a strong foundation in mathematical modeling, machine learning, and biomedical data science, and aims to apply these skills to improve outcomes after lung transplantation. Through this K25 award, he will transition to an independent research career centered on translational, data-driven solutions to clinical challenges in transplant medicine. This will be achieved under the following Aims: Aim 1: Identify and validate predictors of PGD incidence and severity. Aim 2: Identify and validate predictors of CLAD onset and subtype. In pursuit of both Aims, we will obtain biospecimens from 3 anatomical sites (donor lung pre-transplant, donor lung 2 hrs post-transplant, and serially blood samples from the recipient) and donor and recipient clinical data from ≥200 patients at two large lung transplant centers. We will perform a multiplex assay measuring 71 cytokine/growth factors to screen samples for markers of cell activation, chemotaxis, injury, angiogenesis and growth factors. We will use computational phenotyping and machine learning to predict outcomes from these comprehensive biological and clinical data. These models will also allow us to identify novel markers for both outcomes that could improve mechanistic understanding and support point-of-care test design. Trained models and candidate markers will be validated using multiplex and ELISA assays of prospectively collected samples from recipients at two major transplant centers over the course of the study. The ability to predict PGD and to forecast CLAD onset are essential for the development of life-, health-, and graft-extending interventions. Our proposed project builds upon important ongoing work while introducing five advances: multi-site specimens, a broad-scope injury/immune response cytokine panel, analysis of conventional storage solution, specialized computational modeling, and multi-center validation. Successful completion of this project will yield externally validated biomarker panels for PGD incidence and severity and CLAD onset and subtype suitable for developing point-of-care tests.
Relevance Statement
Lung transplant recipients face worse outcomes than those of other solid organs largely due to the frequent and unpredictable onset of primary graft dysfunction (PGD) and chronic lung allograft dysfunction (CLAD). The proposed project aims to improve early identification of high-risk patients through machine learning on clinical and biological data, which will inform the development of predictive tools that enable earlier intervention to improve survival. This award will support the candidate’s transition to an independent research career focused on computational solutions to urgent problems in lung transplantation.