Predicting Gene Disease Associations in Type 2 Diabetes Using Machine Learning on Single-Cell RNA-Seq Data

Diabetes is a chronic metabolic disorder characterized by elevated blood glucose levels due to impaired insulin production or function. Two main forms are recognized: type 1 diabetes (T1D), which involves autoimmune destruction of insulin-producing \b{eta}-cells, and type 2 diabetes (T2D), which arises from insulin resistance and progressive \b{eta}-cell dysfunction. Understanding the molecular mechanisms underlying these diseases is essential for the development of improved therapeutic strategies, particularly those targeting \b{eta}-cell dysfunction. To investigate these mechanisms in a controlled and biologically interpretable setting, mouse models have played a central role in diabetes research. Owing to their genetic and physiological similarity to humans, together with the ability to precisely manipulate their genome, mice enable detailed investigation of disease progression and gene function. In particular, mouse models have provided critical insights into \b{eta}-cell development, cellular heterogeneity, and functional failure under diabetic conditions. Building on these experimental advances, this study applies machine learning methods to single-cell transcriptomic data from mouse pancreatic islets. Specifically, we evaluate two supervised approaches identified in the literature; Extra Trees Classifier (ETC) and Partial Least Squares Discriminant Analysis (PLS-DA), to assess their ability to identify T2D-associated gene expression signatures at single-cell resolution. Model performance is evaluated using standard classification metrics, with an emphasis on interpretability and biological relevance

💡 Research Summary

This study investigates the utility of supervised machine‑learning methods for distinguishing diabetic from non‑diabetic pancreatic β‑cells using single‑cell RNA‑sequencing data from mouse models. The authors leveraged the Mouse Islet Atlas, a curated resource that integrates over 300,000 cells from multiple studies, and focused on two well‑characterized type‑2 diabetes models: the db/db leptin‑receptor mutant and the streptozotocin‑treated (mSTZ) model. After extracting only β‑cells, they performed standard preprocessing steps—quality filtering, CPM normalization, log‑transformation, and selection of highly variable genes (≈2,000). Batch effects across the integrated datasets were mitigated using Harmony, and class imbalance was addressed through a combination of SMOTE oversampling and undersampling.

Two supervised classifiers were evaluated: an Extra Trees Classifier (ETC), an ensemble of randomized decision trees that provides intrinsic feature‑importance scores, and Partial Least Squares Discriminant Analysis (PLS‑DA), a dimensionality‑reduction technique that maximizes covariance between gene expression and the disease label. Both models were trained and validated using five‑fold cross‑validation, and performance was quantified with accuracy, precision, recall, F1‑score, and ROC‑AUC. The ETC achieved an average accuracy of 84 % and an AUC of 0.91, while PLS‑DA reached an accuracy of 81 % and an AUC of 0.89. Feature‑importance analysis from the ETC highlighted a set of ~20 genes involved in insulin signaling, oxidative stress, and cell‑cycle regulation. PLS‑DA latent‑space loadings identified ~15 genes that contributed most strongly to class separation, many of which are linked to β‑cell dedifferentiation and metabolic reprogramming. Notably, several genes uncovered by both approaches overlap with known human T2D biomarkers, and a few novel candidates (e.g., Xist, Gadd45g) were proposed for further experimental validation.

The authors discuss several limitations: reliance on mouse data limits direct translatability to human disease, and the absence of multi‑omics integration may overlook epigenetic or proteomic contributors. They suggest future work should incorporate human single‑cell datasets, apply advanced interpretability tools such as SHAP or LIME, and explore deep‑learning architectures that can capture nonlinear gene‑gene interactions. Overall, the paper demonstrates that relatively simple, interpretable machine‑learning models can achieve high predictive performance on high‑dimensional single‑cell transcriptomic data and can generate biologically meaningful gene signatures relevant to type‑2 diabetes pathogenesis.