Body-centered cubic (bcc) Fe-Mn systems are known to exhibit a complex and atypical magnetic behaviour from both experiments and 0 K electronic-structure calculations, which is due to the half-filled 3d-band of Mn. We propose effective interaction models for these alloys, which contain both atomic spin and chemical variables. They were parameterized on a set of key density functional theory (DFT) data, with the inclusion of non-collinear magnetic configurations being indispensable. Two distinct approaches, namely a knowledge-driven and a machine-learning approach have been employed for the fitting. Employing these models in atomic Monte Carlo simulations enables the prediction of magnetic and thermodynamic properties of the Fe-Mn alloys, and their coupling, as functions of temperature. This includes the decrease of Curie temperature with increasing Mn concentration, the temperature evolution of the mixing enthalpy and its correlation with the alloy magnetization. Also, going beyond the defect-free systems, we determined the binding free energy between a vacancy and a Mn atom, which is a key parameter controlling the atomic transport in Fe-Mn alloys.