Personal prediction of graft survival in kidney transplant recipients before transplantation using machine learning methods

Kidney allograft failure is an important burden in kidney disease and contributes to the increasing number of people with ESKD, now exceeding 7 million worldwide as of 2020. Currently, more than 3000 Kazakhstanis patients urgently need kidney transplantation (www.egov.kz., accessed 28.03.2022). Despite improvements in transplant immune suppression techniques, unadjusted 5-year survival of ESKD patients on KRT was 41% in the USA, 48% in Europe, and 60% in Japan. Recent emerged studies reveal that artificial intelligence-assisted donor and recipient data analysis may improve the prediction of long-term graft survival.

Project goal:
We aim to develop an artificial intelligence approach to enhance risk stratification for kidney transplant recipients of living donors by generating continuously refined predictions of early allograft function impairment and survival before transplantation.

Project Objectives:
Kidney allograft failure is an important burden in kidney disease and contributes to the increasing number of people with ESKD, now exceeding 7 million worldwide as of 2020. Currently, more than 3000 Kazakhstanis patients urgently need kidney transplantation (www.egov.kz., accessed 28.03.2022). Despite improvements in transplant immune suppression techniques, unadjusted 5-year survival of ESKD patients on KRT was 41% in the USA, 48% in Europe, and 60% in Japan. Recent emerged studies reveal that artificial intelligence-assisted donor and recipient data analysis may improve the prediction of long-term graft survival.

Supervisor of the project, Dr. Salybekov Amangeldi

Graduated from Astana Medical University (Kazakhstan) in 2012.
Worked as vascular surgeon at vascular and reconstructive surgery department in JSC ` National Center for Oncology and Transplantation`.
In 2019 received PhD degree in medicine from Tokai University School of Medicine and pursued joint postdoctoral followership in Brigham woman’ hospital, Harvard University School of Medicine and Tokai University School of Medicine.
From 2020, Dr. Salybekov works as CEO of Qazaq Institute of Innovative Medicine and also in Shonan Kamakura General Hospital Kidney Disease and Transplant Center as visiting director.

Scopus: 57204169112
Web of Science: M-1130-2013
ORCID: 0000-0001-5490-9365

Our Team

Timeline of the project

Data collection and preprocessing

Collecting and data preprocessing i.e.

Feature engineering and Predicting models

Feature engineering and Predicting models

Development of predicting models

Algorithms: Lineal regression, clustering, classification, and Apriori models for recipients and donors’ risks stratification and future graft function prediction

Convolutional neural networks classification on allograft biopsies

Three times cross-validation (more than 10000 patients, from USA transplant centers) and deployment on an external real-world cohort (101 patients, Japanese cohort) will be used for validation. The area under the receiver operating characteristic curve (AUROC) will be used as the main performance metric (the primary endpoint to assess CNN performance).

Kidney allograft function assessment and prediction models

Model development. As shown in Fig 2., we will use various models to find the best fit model for the prediction of eGFR.

Currently available prerelease of Kidney pretransplantation predictor app, you can download it by clicking: Predictor

Donor kidney function assessment and predicting models

This step will utilize the Apriori algorithm to define the possible mechanism of transplanted graft functional impairment.

Histological data and genetic data collection and interpretation

Histological and genetic data will be integrated to clinical data to increase eGFR predictive accuracy. Applying various data, we plan to increase sensitivity and specificity.

1st manuscript draft preparation

1st manuscript draft preparation. Reading and collection of literature. Data mining

1st manuscript revision, and submission to the journal

1st article submission to the journal that is included in the 1 (first) and (or) 2 (second) quartile by impact factor in the Web of Science database and (or) having a CiteScore percentile in the Scopus database of at least 65 (sixty-five).

Cloud-based Web-application construction for data analysis

Cloud-based web application first test will start based on preliminary and obtained outcomes.

Data analysis

Final Histology data analysis and re-integration with clinical data.

Clinical samples characteristics

In this step, we use genetic data to define possible mechanism of early transplant functional mitigation via machine learning algorithms.

Published articles review and data collection, mining

The 2nd manuscript draft preparation, revision.

Final analyzed data evaluation

Analyzed data models will be verified and checked.

2nd manuscript final revision and submission

2nd article submission to the journal that is included in the 1 (first) and (or) 2 (second) quartile by impact factor in the Web of Science database and (or) having a CiteScore percentile in the Scopus database of at least 65 (sixty-five).

Final cloud-based web-application

Final cloud-based web-application.

Ask about project

General:

amansaab0@gmail.com, markus.wolfien@gmail.com


Biopsy image processing:

zhburibaev@gmail.com


Tabular data analysis:

aidyn.daulet@gmail.com, yerkosova@gmail.com