Hydrocarbon pyrolysis is a complex chemical reaction system under extreme temperature and pressure conditions, involving numerous chemical reactions and chemical species. Only two kinds of atoms are involved: carbons and hydrogens. Its effective description and prediction in new settings are challenging due to the system's complexity and the high computational cost of generating data via molecular dynamics simulations. However, the ensemble of molecules present at any moment, and the carbon skeletons of these molecules, can be viewed as random graphs. This approach has the advantage over the traditional kinetic Monte Carlo: it has a low computational cost, requires only a few reaction constant pairs to be learned from molecular dynamics data, and yields results completely determined by the H/C ratio and temperature at a fixed pressure. The lower the H/C ratio, the more complex the random graph model must be to accurately predict the equilibrium molecular size distribution and the giant component. I will describe our best “disjoint loops” model (DOI: 10.1103/PhysRevE.111.034303), which features two motifs: edges and small loops. This model is adequate for reaction systems at a pressure of 40.5 GPa, a temperature range of 3200–5000 K, and an H/C ratio ranging from 2.25 (as in octane) to 4 (as in methane). However, as the H/C decreases further, loops begin to form clusters, which effectively reduces the size of the giant component. I will describe our further developments to address this case.