Towards a brand ney way to understant kidney cancer: un unsupervised machine learnign approach

ROMEO, LEONARDO RAFAEL; NÚÑEZ, MATIAS; FERRANDO, MATÍAS; LOPEZ-FONTANA, CONSTANZA MATILDE; CARÓN, RUBÉN WALTER; BRUNA, FLAVIA ALEJANDRA; PISTONE-CREYDT, VIRGINIA

Mar del Plata

Congreso; REUNIÓN ANUAL DE SOCIEDADES DE BIOCIENCIAS; 2022

Among the different types of cells that surround renal epithelialcells, renal adipose tissue (AT) is one of the most abundant. Wedemonstrated that human renal adipose tissue from patients withrenal tumors (hRAT) regulates the behavior of epithelial cells differently from normal renal adipose tissue (hRAN), through the proteins expression characterization in hRAT vs hRAN. In this work,we evaluated: 1) the differential proteins expression as a whole wassufficient to separate healthy patients from patients with kidney cancer, using unsupervised machine learning algorithms (UMLA); 2) thecorrelation between adiponectin and leptin expression with clinicalcharacteristics of kidney cancer patients. The biological variablesevaluated in Hrat (n=21) and hRAN (n=24) were: adiponectin, AdipoR, leptin, ObR, perilipin and ADAMTS1. The proteins expressionby the different ATs were analyzed with UMLA algorithms (t-SNEand UMAP). We selected leptin and adiponectin to study the correlation with clinical characteristics of patients with kidney tumors(sex, age, BMI, smoking, tumor grade, size of the lesion, densityof AT, and difficulty in the surgical dissection). The SPSS programwas used to statystical analysis, taking a significant p < 0.05. Considering the total of biological variables evaluated in the differentAT fragments, we were able to separate healthy from kidney tumorpatients by UMLA projection. A decrease in adiponectin expressionwas found in patients with a more undifferentiated tumor related withsmoking habits (p