Machine learning applied to gene expression analysis of T-lymphocytes in patients with cGVHD

Machine learning-inspired approaches allow for processing of multidimensional data and have shown to uncover clinical patterns in several human diseases. Recently it has been used to model cGVHD risk groups according to clinical data, but the role of this approach applied to gene-expression-profile (GEP) data has not been established yet. In this study, we aimed to unravel specific GEP patterns associated to cGVHD using machine-learning analyses. For this purpose, we isolated peripheral blood T-lymphocytes drawn from normal healthy donors and from patients who had underwent allogeneic hematopoietic stem cell transplantation (allo-HSCT) with and without cGVHD symptoms. Dataset obtained in HG-U133 Plus 2.0 Gene Chip oligonucleotide array were wrapped by Boruta method based on Random Forest. Thus, a limited set of 53 genes was retained and two-dimensional Principal Component Analysis plots projection was plotted showing a clear distinction of cGVHD with No-cGVHD and healthy Control groups with area under the curve (AUC) over 0.75 for each comparison. The highest scored genes were CDKN2A, SERPINB9, LYPLA1 y CKTM1A/B genes which are involved in positive regulation of cellular senescense with protection against perforin-dependent apoptosis. Altogether our novel findings, using machine-learning approach applied to GEP in cGVHD unravel a limited panel of five genes with a potential diagnostic and targeted therapy.