Nanosensor for mass detection is a mechanical sensor that from an experimental and theoretical point of view proved to be an excellent candidate in the detection of atoms and molecules. The basic idea of these sensors is the frequency shift method. This method is based on the dierence between the resonant frequency of graphene with and without added mass in the form of atoms and molecules bonded with carbon atoms. Natural frequencies, and the frequency shift method used in atom/molecule detection will be analyzed using molecular dynamics (MD) simulation and non-local theory of plates. Numerical results obtained using MD (natural frequencies) will be used to adjust the nonlocal parameter value in non-local thin plate theory in such a way that the frequency from non-local thin plate theories will be equated with the natural frequencies from MD by modifying the non-local parameter value. This will create a dataset for the implementation of the genetic programming algorithm in order to establish approximate correlations between the input data (mechanical characteristics of graphene, graphene dimensions, temperature and natural frequencies) with the output data (non-local parameter). Since most sensors have a specied temperature range and pressure at which the sensor can perform detection, it was initially assumed that graphene as a sensing element of the nanosensor can detect gas molecules in the range from 233.15 to 313.15 K and at a pressure in the range from 0 to 1 bar. Before investigating the natural frequencies of one layer of graphene, the mechanical and thermodynamic characteristics of this material were obtained using MD with REBO interatomic potential. The obtained mechanical and thermodynamic parameters were used in non-local thin plate theory to determine the natural frequencies of single-layer graphene as well as to examine the inuence of temperature, pressure, size of single-layer graphene, a variation of non-local graphene parameter on natural frequencies caused by gas molecules attached to the surface of the single-layer graphene. MD was used to determine the natural frequencies of single-layer graphene, to examine the inuence of single-layer graphene size on natural frequencies as well as the absolute and relative frequency shift caused by attached gas molecules to the central single-layer graphene atom using the displacement excitation method. The results of the above analyzes in MD and non-local thin plate theory showed that the size of single-layer graphene has the greatest inuence on the natural frequencies of single-layer graphene while temperature has a very small inuence on the natural frequencies. MD simulations with NPT ensemble showed that the pressure oscillates a lot during equilibration and vibration simulation and therefore the inuence of pressure was omitted from further analyzes. Analyzes performed using non-local thin plate theory also showed that the value of the non-local parameter has a large inuence on the natural frequencies of single-layer graphene. The mechanical characteristics are temperature dependent and with increasing temperature, the value of these parameters with small oscillations gradually decreases. Physical data of 3 molecules of chemical weapons of mass destruction were used to investigate the possibility of detecting gas molecules using non-local theory and MD. Both theories have shown that single-layer graphene can detect gas molecules using the absolute and relative frequency shift method. One of the main shortcomings of the non-local theory of elasticity is the unknown value of the non-local parameter, and in many studies, its value is set in a certain range. Therefore, the goal is to apply a genetic programming algorithm to obtain a symbolic expression by which its value could be determined. The genetic programming algorithm was used to determine the symbolic expression that would connect the input values with mathematical functions, namely: mechanical parameters (modulus of elasticity, shear modulus, Poisson's coecients and dimensions of graphene), operating conditions (temperature) and natural frequencies obtained by MD with a non-local parameter that represents the output value of this symbolic expression. For a genetic programming algorithm to be able to generate a symbolic expression, it is necessary to develop a data set on which the algorithm can be trained and tested. Three temperatures (233.15, 273.15, and 313.15 K) were used to generate the data set in MD and non-local theory. Graphene models with dimensions ranging from 20× 10 to 40×20 nm were used for these simulations. The natural frequencies obtained by MD were used as reference values for tuning the natural frequencies in non-local theory so that the value of the non-local parameter was adjusted so that the natural frequency value was equivalent to those frequencies obtained in MD. Based on the obtained parameters using MD and non-local theory, a set of data was created which was used in the genetic programming algorithm to determine the equation by which the value of the non-local parameter could be determined. The equation for determining the value of a non-local parameter was chosen based on the highest achieved value of the R2 correlation coecient which is equal to 0.9688. The obtained equation for determining the non-local parameter value was used to determine the absolute and relative frequency shift caused by the mass attached in the center of the fully clamped graphene sheet. The obtained results were compared with the averaged absolute and relative frequency shift values obtained using MD simulations. The results of the comparison showed that on average the calculated values of absolute and relative frequency shift are 5 % lower than those obtained using MD simulations.