Machine learning methodology using a masked neural network for robust genetic risk score calculation from noisy and missing data.

Masked-multi-layer perceptron improves robustness of genetic risk scores, achieving a Spearman correlation of 0.951 with noisy data compared to 0.669 for traditional methods.

• Why it matters: Accurate genetic risk scores are crucial for disease prediction but are often compromised by noisy or missing data, limiting their clinical utility.
• What they did: A neural network model called masked-MLP was developed, trained on clean data with added noise and extra inputs, to produce more reliable GRS estimates under data imperfections.
• The result: The masked-MLP significantly outperformed standard approaches in noisy conditions, enabling more dependable GRS calculations and potentially improving disease risk assessments.