Reverse Design of Biodegradable Mg Alloys Using Interpretable Machine Learning and Genetic Algorithm

Researchers from the Korea Institute of Materials Science have proposed a novel approach to design biodegradable Mg-Zn-Mn-Sr-Ca (ZMJX) alloys for load-bearing orthopedic implants. By utilizing interpretable machine learning and a multi-objective genetic algorithm, the team aimed to optimize the composition of the alloys to achieve both mechanical strength and biologically compatible degradation rates. According to the study, the optimized neural network models predicted the ultimate compressive strength and in vitro degradation rate with high accuracy, exceeding 0.92 on testing data.

Key Takeaways:

• The researchers developed a reverse engineering approach that used interpretable machine learning and a multi-objective genetic algorithm to design biodegradable Mg-Zn-Mn-Sr-Ca alloys for load-bearing orthopedic implants.
• The training of neural networks was facilitated by a dataset comprising 1044 data points, which contained chemical compositions, ultimate compressive strengths (UCS), and in vitro degradation rates (DR).
• Shapley additive explanations identified Zn as the most influential element affecting both UCS and DR.
• A total of six optimal alloys were identified from the Pareto front of two fitness functions (UCS 240 MPa, DR 1 mm/y), four of which satisfied the conflicting dual target criteria.
• Notable performance discrepancies were observed between reverse design and experimental results.
• The research emphasizes the importance of incorporating microstructural and textural characteristics into existing databases to enhance model accuracy and enable more reliable, data-driven design of biodegradable Mg alloys.
• The study was supported by the Korea Institute of Materials Science, National Research Foundation of Korea, and Korea Ministry of Science And Ict.

Statistics:

• 1044 data points were used to train the neural networks.
• The optimized neural network models predicted UCS and DR with a coefficient of determination exceeding 0.92 on testing data.
• 6 optimal alloys were identified from the Pareto front of two fitness functions.
• 4 of the identified alloys satisfied the conflicting dual target criteria.
• 240 MPa was the target ultimate compressive strength (UCS) for the alloys.
• 1 mm/y was the target in vitro degradation rate (DR) for the alloys.

Sources:

• Reverse design of Mg-Zn-Mn-Sr-Ca alloys for biodegradable implants by interpretable machine learning and genetic algorithm, Materials & Design, 2025,257():114494. (Elsevier)
• Korea Institute of Materials Science.