Predicting Protein and Fat Content in Human Donor Milk Using Machine Learning.

Area of Science:

• Nutritional Science
• Biotechnology
• Data Science

Background:

• Donor milk is crucial for hospitalized very low birth weight (VLBW) infants when mother's milk is unavailable, but its variable nutrient composition often leads to suboptimal infant growth.
• Standardizing donor milk composition is challenging due to variations in individual donations, impacting the provision of adequate nutrition for VLBW infants.

Purpose of the Study:

• To evaluate the accuracy of machine learning models in predicting the macronutrient content (specifically fat and protein) of donor milk.
• To determine if machine learning can help reduce nutrient variability in donor milk, thereby supporting optimal growth and development in VLBW infants.

Main Methods:

• Collected 272 individual and 61 pooled donor milk samples from a human milk bank for macronutrient analysis.
• Developed four machine learning models using donation, donor, and pumping practice variables, with lactation stage and infant gestational status as a baseline.
• Predicted macronutrient content for both individual donations and pooled samples.

Main Results:

• Machine learning models achieved mean absolute errors of 0.16 g/dL for individual donation protein and 0.10 g/dL for pooled protein.
• Fat prediction MAEs were 0.91 g/dL for individual donations and 0.42 g/dL for pools.
• Protein predictions were significantly more accurate than the baseline, while fat predictions were competitive with the baseline.

Conclusions:

• Machine learning models demonstrate high accuracy in predicting donor milk macronutrient content.
• Pooling donor milk samples reduces prediction error, highlighting the importance of current milk banking practices.
• Future research should focus on integrating these predictive models into milk bank pooling strategies to enhance nutritional delivery for VLBW infants.