|
|
 |
| |
|
|
|
Optimized Bi-clustering Framework for AML Biomarker Discovery: Integrating PHATE-Based Dimensionality Reduction and Enhanced Gene Expression Clustering |
|
|
|
PP: 763-775 |
|
|
doi:10.18576/amis/200315
|
|
|
|
Author(s) |
|
|
|
Tarik Himdi,
Mohammed Ishaque,
Khaled ElBahnasy,
|
|
|
|
Abstract |
|
|
| One of the key features of AML is genetic heterogeneity, which poses major problems in developing molecular markers that could be employed to improve diagnostic and individualized therapy accuracy. In this study, we proposed a methodological framework that uses a refined version of the biclustering algorithm of Cheng and Church. The novelty of our approach consists in the use of advanced techniques for reducing the dimension of the data, especially the use of PHATE, in addition to UMAP and NMF methods, to optimize the biclustering analysis. We identified 61 significant biclusters, including 66 critical genes. They include classical AML markers, such as CD38, MPO, and CREBBP, as well as potentially new genes, PLXND1 and ALS2, that might indicate additional molecular mechanisms of disease formation. All identified biclusters were validated according to their biological significance and statistical reliability using silhouette scoring, cluster coherence score, and comparison with the current scientific literature. Our proposed approach provides better clustering than classical approaches such as PCA based on silhouette scores. |
|
|
|
|
|
|
|
