Within the field of High Energy Physics various machine learning techniques are being constantly researched. Due to upcoming HL-LHC (High Luminosity Large Hadron Collider) upgrade, which aims to increase probability of occurrence of rare physical phenomenas by significant increase of number of collisions occurring each second, currently used algorithms in the area of charged particle reconstructions are not going to scale well enough. The data from tracking detectors if the LHC experiments can be represented in the form of a graph, where detected signals are represented as nodes and hypothetical connections between them are represented by edges. In particular the initial stage of track finding, the formation of seeds can be represented in that way. Within this paper we propose solution based on Graph Neural Network used to dramatically decrease number of seeds and therefore significantly reduce amount of calculations needed by downstream algorithms in order to reconstruct tracks. The numerical experiments were carried out with use a data set obtained from Monte Carlo simulation of the Open Data Detector. The initial results showed, that proposed solution based on Graph Neural Network has application potential.