CD-NuSS

CD-NuSS predicts nucleic acid secondary structures from circular dichroism (CD) spectral data using machine learning.

Key Features:

• Input data: Circular dichroism (CD) spectra as the basis for secondary-structure prediction.
• Algorithms: Extreme gradient boosting decision-tree (XGBoost) and neural network (nnet) models for classification.
• Training dataset: A curated library of 450 CD spectra representing 16 distinct nucleic acid secondary structures.
• Hyper-parameter optimization: Optimization performed using holdout and k-fold cross-validation with k = 5.
• Validation: Tested on a dataset of 150 CD spectra with reported prediction accuracies of 85%–87%, with XGBoost achieving slightly higher accuracy.
• Predictive scope: Capable of predicting hybrid nucleic acid topologies.

Scientific Applications:

• Secondary-structure quantification: Estimating quantitative secondary structural content of nucleic acids from CD spectral data.
• Topology identification: Predicting canonical and hybrid nucleic acid topologies from CD spectra.
• Structural monitoring: Supporting analysis of structural changes during biomolecular interactions using CD data.

Methodology:

XGBoost and nnet models were trained on 450 curated CD spectra (16 structure classes), hyper-parameters were tuned via holdout and 5-fold cross-validation, and models were evaluated on a 150-spectrum test set (85%–87% accuracy, XGBoost higher).