Machine learning analysis of 152 heart attack patients revealed that just three biomarkers can achieve 94% accuracy in distinguishing severe STEMI from NSTEMI heart attacks, with matrix metalloproteinase-2 (MMP-2) serving as the dominant predictor. The biomarker MMP-2 alone contributed approximately 16% to model performance, while MMP-2 combined with EMMPRIN showed synergistic effects exceeding their individual contributions. This structured approach to understanding AI model architecture represents a significant methodological advance beyond simply measuring overall prediction accuracy. The findings suggest that expensive multi-biomarker panels currently used in clinical practice could potentially be streamlined to focus on just a few high-yield markers, reducing costs while maintaining diagnostic precision. However, this preprint study awaits peer review, and the relatively small sample size of 152 patients limits generalizability. The work addresses a critical gap in cardiovascular AI research by examining not just whether machine learning works, but how it works at the molecular level. If validated in larger cohorts, this approach could reshape biomarker selection strategies for post-heart attack risk stratification and guide more targeted therapeutic interventions.