Abstract
Obstructive Nephropathy (ON) is a renal disease and its pathology is believed to be magnified by various molecular processes. In the current study, we apply an intelligent workflow implemented in Rapidminer data mining platform to two different ON datasets. Our scope is to select the most important actors in two corresponding molecular information levels: human miRNA and mouse mRNA. A forward selection method with an embedded nearest neighbor classifier is initially applied to select the most important features in each level. The resulting features are next fed to classifiers appropriately tested utilizing a leave-one-out resampling technique in order to evaluate the relevance of the selected input features when used to classify subjects into output classes defined by ON severity. Preliminary results show that high classification accuracies are obtained, and are supported by the fact that the selected miRNAs or mRNAs have been found significant within differential expression analysis using the same datasets.
Introduction
Obstructive Nephropathy (ON) occurs when the flow of urine is blocked, leading to inflammation, cell death, and eventually permanent kidney scarring (fibrosis). While the mechanical cause is clear, the underlying molecular "magnifiers" that drive the damage are complex and not fully understood. Identifying these biomarkers is complicated by the "curse of dimensionality": modern microarray experiments measure tens of thousands of genes (mRNAs) or regulatory molecules (miRNAs) simultaneously, but researchers often only have access to a small number of patient samples. Finding specific biomarkers that predict disease progression in this "small $n$, large $p$" scenario requires sophisticated computational tools to avoid finding patterns that are merely statistical noise.
Methodology
The authors implemented a multi-stage intelligent workflow using the Rapidminer data mining platform. The process involves several key steps:
- Feature Selection (Forward Selection): Instead of looking at all genes at once, the algorithm starts with an empty set and adds one feature at a time. In each round, it adds the feature that most improves the performance of an embedded 6-nearest neighbor (6-nn) classifier. This process continues until no further improvement is seen or a limit (30 features) is reached.
- Classification & Validation: Once the best subset of features was identified, the researchers tested them using several classifiers: 1-nn, 6-nn, 12-nn, and a Classification Tree.
- Resampling: To ensure the results were robust despite the small sample size, they used leave-one-out (LOO) cross-validation, where the model is repeatedly trained on all but one sample and tested on the remaining one.
Datasets
The study utilized two distinct "omics" datasets:
- Human miRNA Dataset: Samples from 26 children (aged 2 weeks to 6 months) categorized into three groups: Control (8), Mild/No surgery needed (8), and Severe/Surgery needed (10). It tracked 790 different miRNAs.
- Mouse mRNA Dataset: Samples from 18 neonatal mice following a procedure to mimic human ON, categorized as Control (9), Mild (5), and Severe (4). This dataset was much larger, covering ~41,000 transcripts corresponding to 20,873 genes.
Results and Discussion
The workflow proved highly effective at identifying a "minimalist" set of biomarkers across both species. Interestingly, the features selected by the machine learning algorithm weren't always the ones with the absolute lowest p-values in traditional statistics. However, they provided the best predictive power when used together, suggesting that the "intelligent" selection captures synergistic relationships between markers.
Human miRNA Results
The model selected 7 miRNAs (including kshv-miR-K12-9 and hsa-miR-125b). These seven markers achieved classification accuracies between 82% and 93%* across different testing models.
Mouse mRNA Results
The model narrowed 20,000+ genes down to just 3 key mRNAs: Tpm1 (tropomyosin 1), Svs1 (seminal vesicle secretory protein 1), and Wnt4 (a well-known signaling protein in kidney development). These three genes achieved a staggering 100% accuracy in some classifier configurations.
Conclusion
This research demonstrates that intelligent feature selection can successfully identify critical molecular players in Obstructive Nephropathy. By identifying markers in both humans and mice, the study sets the stage for translational research to validate mouse models for human drug trials. Future work will involve more complex models like Artificial Neural Networks and investigating if the human miRNAs selected actually regulate the specific mouse mRNAs identified, potentially revealing a complete regulatory circuit for the disease.
Summary updated