Prediction of the development of islet autoantibodies through integration of environmental, genetic, and metabolic markers.

BACKGROUND: The Environmental Determinants of the Diabetes in the Young (TEDDY) study has prospectively followed, from birth, children at increased genetic risk of type 1 diabetes. TEDDY has collected heterogenous data longitudinally to gain insights into the environmental and biological mechanisms driving the progression to persistent islet autoantibodies. METHODS: We developed a machine learning model to predict imminent transition to the development of persistent islet autoantibodies based on time-varying metabolomics data integrated with time-invariant risk factors (eg, gestational age). The machine learning was initiated with 221 potential features (85 genetic, 5 environmental, 131 metabolomic) and an ensemble-based feature evaluation was utilized to identify a small set of predictive features that can be interrogated to better understand the pathogenesis leading up to persistent islet autoimmunity. RESULTS: The final integrative machine learning model included 42 disparate features, returning a cross-validated receiver operating characteristic area under the curve (AUC) of 0.74 and an AUC of ~0.65 on an independent validation dataset. The model identified a principal set of 20 time-invariant markers, including 18 genetic markers (16 single nucleotide polymorphisms [SNPs] and two HLA-DR genotypes) and two demographic markers (gestational age and exposure to a prebiotic formula). Integration with the metabolome identified 22 supplemental metabolites and lipids, including adipic acid and ceramide d42:0, that predicted development of islet autoantibodies. CONCLUSIONS: The majority (86%) of metabolites that predicted development of islet autoantibodies belonged to three pathways: lipid oxidation, phospholipase A2 signaling, and pentose phosphate, suggesting that these metabolic processes may play a role in triggering islet autoimmunity.

Better understanding and prediction of antiviral peptides through primary and secondary structure feature importance.

The emergence of viral epidemics throughout the world is of concern due to the scarcity of available effective antiviral therapeutics. The discovery of new antiviral therapies is imperative to address this challenge, and antiviral peptides (AVPs) represent a valuable resource for the development of novel therapies to combat viral infection. We present a new machine learning model to distinguish AVPs from non-AVPs using the most informative features derived from the physicochemical and structural properties of their amino acid sequences. To focus on those features that are most likely to contribute to antiviral performance, we filter potential features based on their importance for classification. These feature selection analyses suggest that secondary structure is the most important peptide sequence feature for predicting AVPs. Our Feature-Informed Reduced Machine Learning for Antiviral Peptide Prediction (FIRM-AVP) approach achieves a higher accuracy than either the model with all features or current state-of-the-art single classifiers. Understanding the features that are associated with AVP activity is a core need to identify and design new AVPs in novel systems. The FIRM-AVP code and standalone software package are available at https://github.com/pmartR/FIRM-AVP with an accompanying web application at https://msc-viz.emsl.pnnl.gov/AVPR .

Bayesian Model Averaging for Ensemble-Based Estimates of Solvation-Free Energies.

This paper applies the Bayesian Model Averaging statistical ensemble technique to estimate small molecule solvation free energies. There is a wide range of methods available for predicting solvation free energies, ranging from empirical statistical models to ab initio quantum mechanical approaches. Each of these methods is based on a set of conceptual assumptions that can affect predictive accuracy and transferability. Using an iterative statistical process, we have selected and combined solvation energy estimates using an ensemble of 17 diverse methods from the fourth Statistical Assessment of Modeling of Proteins and Ligands (SAMPL) blind prediction study to form a single, aggregated solvation energy estimate. Methods that possess minimal or redundant information are pruned from the ensemble and the evaluation process repeats until aggregate predictive performance can no longer be improved. We show that this process results in a final aggregate estimate that outperforms all individual methods by reducing estimate errors by as much as 91% to 1.2 kcal mol-1 accuracy. This work provides a new approach for accurate solvation free energy prediction and lays the foundation for future work on aggregate models that can balance computational cost with prediction accuracy.