CODI: Enhancing machine learning-based molecular profiling through contextual out-of-distribution integration

Molecular analytics increasingly utilize machine learning (ML) for predictive modeling based on data acquired through molecular profiling technologies. However, developing robust models that accurately capture physiological phenotypes is challenged by a multitude of factors. These include the dynamics inherent to biological systems, variability stemming from analytical procedures, and the resource-intensive nature of obtaining sufficiently representative datasets. Here, we propose and evaluate a new method: Contextual Out-of-Distribution Integration (CODI). Based on experimental observations, CODI generates synthetic data that integrate unrepresented sources of variation encountered in real-world applications into a given molecular fingerprint dataset. By augmenting a dataset with out-of-distribution variance, CODI enables an ML model to better generalize to samples beyond the initial training data. Using three independent longitudinal clinical studies and a case-control study, we demonstrate CODIs application to several classification scenarios involving vibrational spectroscopy of human blood. We showcase our approachs ability to enable personalized fingerprinting for multi-year longitudinal molecular monitoring and enhance the robustness of trained ML models for improved disease detection. Our comparative analyses revealed that incorporating CODI into the classification workflow consistently led to significantly improved classification accuracy while minimizing the requirement of collecting extensive experimental observations. SIGNIFICANCE STATEMENTAnalyzing molecular fingerprint data is challenging due to multiple sources of biological and analytical variability. This variability hinders the capacity to collect sufficiently large and representative datasets that encompass realistic data distributions. Consequently, the development of machine learning models that generalize to unseen, independently collected samples is often compromised. Here, we introduce CODI, a versatile framework that enhances traditional classifier training methodologies. CODI is a general framework that incorporates information about possible out-of-distribution variations into a given training dataset, augmenting it with simulated samples that better capture the true distribution of the data. This allows the classification to achieve improved predictive performance on samples beyond the original distribution of the training data.