ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: PO2178

Machine Learning and Bioinformatics Approaches to Discover Urine Gene Expression Biomarkers for Kidney Transplant Rejection

Session Information

Category: Transplantation

  • 1902 Transplantation: Clinical

Authors

  • Kandpal, Manoj, Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
  • Keslar, Karen S., Cleveland Clinic Lerner Research Institute, Cleveland, Ohio, United States
  • Friedewald, John J., Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
  • Fairchild, Robert L., Cleveland Clinic Lerner Research Institute, Cleveland, Ohio, United States

Group or Team Name

  • CTOT Consortium
Background

While transplant biopsies are safe and accurate way for monitoring transplant progress, they are associated with defined risks and significant costs. Using the mRNA from urine, we proposed a non-invasive approach to diagnose acute rejection. The multi-step approach involved collection and mRNA analysis of urine samples, application of a machine learning algorithm to select an initial set of gene markers, adding known markers from prior work, and using the combined set for developing a final classifier. The classifier was developed on a training data consisting of 42 samples (17 rejection and 15 control) and a validation data set of 43 samples (13 rejection and 30 control).

Methods

RNA from urine samples was hybridized to customized NanoString panel, consisting of 796-gene Immune Profiling gene panel and 26 genes representing graft rejection or the development of fibrosis from published works. The RNA samples were processed on the nCounter GEN2 using the high sensitivity protocol and high-resolution data capture. Raw data were imported into nSolver4.0 (NanoString) followed by log2 gene counts and normalized data generation that was used in further analyses.

Results

Using Random Forest on NanoString data we first obtained a set of eight gene as our initial pool of markers. We combined them with the 20 gene markers from our previous work and developed a seventeen gene classifier (after removing duplicates and non-important genes). The combined signature of 17 genes had high AUC (0.875), accuracy (0.881), sensitivity (0.765), specificity (0.96), PPV (0.929) and NPV (0.857) on training data. Although the PPV dipped to 0.714 in the validation data, it still performed well resulting in high accuracy (0.84), sensitivity (0.77), specificity (0.87) and NPV (0.90).

Conclusion

This initial hybrid modeling approach has shown its significance and we plan to further strengthen its reliability and test robustness by incorporating more samples. The final classifier, a non-invasive approach to classify kidney graft heath, could help improve serial monitoring of graft recipients while reducing the cost and safety risks associated with biopsies.

Funding

  • Other NIH Support