New algorithm cuts time and guesswork out of analyzing complex biological data
Researchers have developed a faster way to sort through large, multifaceted datasets common in drug discovery and clinical diagnostics. The approach automates what previously required researchers to manually build and compare dozens of separate analyses, potentially accelerating how quickly biotech companies and hospitals can identify disease signatures or drug candidates.
Originaltitel: OPLS-based multiclass classification and data-driven interclass relationship discovery
<p>Multiclass data sets and large-scale studies are increasingly common in omics sciences, drug discovery, and clinical research due to advancements in analytical platforms. Efficiently handling these data sets and discerning subtle differences across multiple classes remains a significant challenge. In metabolomics, two-class orthogonal projection to latent structures discriminant analysis (OPLS-DA) models are widely used due to their strong discrimination capabilities and ability to provide interpretable information on class differences. However, these models face challenges in multiclass settings. A common solution is to transform the multiclass comparison into multiple two-class comparisons, which, while more effective than a global multiclass OPLS-DA model, unfortunately results in a manual, time-consuming model-building process with complicated interpretation. Here, we introduce an extension of OPLS-DA for data-driven multiclass classification: orthogonal partial least squares-hierarchical discriminant analysis (OPLS-HDA). OPLS-HDA integrates hierarchical cluster analysis (HCA) with the OPLS-DA framework to create a decision tree, addressing multiclass classification challenges and providing intuitive visualization of interclass relationships. To avoid overfitting and ensure reliable predictions, we use cross-validation during model building. Benchmark results show that OPLS-HDA performs competitively across diverse data sets compared to eight established methods. This method represents a significant advancement, offering a powerful tool to dissect complex multiclass data sets. With its versatility, interpretability, and ease of use, OPLS-HDA is an efficient approach to multiclass data analysis applicable across various fields.</p>