The substitution of natural gas hydrates with CO2 offers a compelling dual advantage by enabling the extracting of CH4 while simultaneously sequestering CO2. This process, however, is intricately tied to the mechanical stability of CO2-CH4 heterohydrates. In this study, we report the mechanical properties and cage transformations in CO2-CH4 heterohydrates subjected to uniaxial straining via molecular dynamics (MD) simulations and machine learning (ML). Results indicate that guest molecule occupancy, the ratio of CO2 to CH4 and their spatial arrangements within heterohydrate structure greatly dictate the mechanical properties of CO2-CH4 heterohydrates including Young’s modulus, tensile strength, and critical strain. Notable, the introduction of CO2 within clathrate cages, particularly within 512 small cages, weakens the stability of CO2-CH4 heterohydrates in terms of mechanical properties. Upon critical strains, unconventional clathrate cages form, contributing to loading stress oscillation before fracture of heterohydrates. Intriguingly, predominant cage transformations, such as 51262 to 4151063 or 425864 and 512 to 425861 cages, are identified, in which 4151062 appears as primary intermediate cage that is able to transform into 4151063, 425862, 425863, 512 and 51262 cages, unveiling the dynamic nature of heterohydrate structures under straining.Additionally, machine learning (ML) models developed using MD data well predict the mechanical properties of heterohydrates, and underscore the critical influence of the spatial arrangement of guest molecules on the mechanical properties. These newly-developed ML models serve as valuable tools for accurately predicting the mechanical properties of heterohydrates. This study provides fresh insights into the mechanical properties and cage transformations in heterohydrates in response to strain, holding significant implications for environmentally sustainable utilization of CO2-CH4 heterohydrates.