A gene selection approach for Diabetic retinopathy microarray data classification using Ant Colony Optimization

Almost all diabetic patients develop diabetic retinopathy. Interpersonal diversity contributes significantly to the susceptibility of serious manifestations of diabetic retinopathy leading to impaired vision. Further, insufficient studies have been performed on the diagnosis of molecular biomarkers for diabetic retinopathy using machine learning. Hence, this study proposed an approach for gene selection in microarray data.

The proposed method involves a primary filter approach that uses the results of different gene expression analyzes, thereby reducing the primary genes and thus the complexity of space and search time. A set of genes that improve classification accuracy are then identified using the Ant Colony Optimization (ACO) method based on the heuristic approach. Selected genes in the final phase are evaluated using a ROC curve (receptor function trait) to determine the most effective while the smallest subset of traits. The classifier evaluated in the proposed framework is the K-nearest neighbor. A set of diabetic retinopathy microarrays is used to test the proposed approach.

The results of the experiment reveal that the our suggested method obtained a high accuracy rate with 9 highly informative genes. Furthermore, we found four genes including ANKDD1A, ZNF786, SNORA3B and, C14orf2 as novel potential molecular biomarker in diabetic retinopathy.

The results showed that the other heuristic algorithms can be used in eye diseases for gene selection. Also, it is worthwhile evaluating them through biological research and experimentation because of the good discrimination power of the selected genes.