Machine Learning
The WEKA data mining software: an update
ACM SIGKDD Explorations Newsletter
| Hi-index | 0.00 |
The elderly population in almost every country is growing faster than ever before. However, our knowledge about the aging process is still limited despite decades of studies on this topic. In this report, we focus on the gradual accumulation of DNA damage in cells, which is a key aspect of the aging process and one that underlies age-dependent functional decline in cells, tissues, and organs. To achieve the goal of discriminating DNA-repair from non-DNA-repair genes among currently known genes related to human aging, four machine learning methods were employed: Decision Trees, Naïve Bayes, Support Vector Machine, and Random Forest (RF). Among the four methods, the RF algorithm achieved a total accuracy (ACC) of 97.32% and an area under receiver operating characteristic (AUC) of 0.98. These estimates were based on 18 selected attributes, including 10 Gene Ontology and 8 Protein-Protein Interaction (PPI) attributes. A predictive model built with only 15 PPI attributes achieved performance levels of ACC= 96.56% and AUC=0.95. Systems biology analyses showed that the features of these attributes were related to cancer, genetic, developmental, and neurological disorders, as well as DNA replication/recombination/repair, cell cycle, cell death, and cell function maintenance. The results of this study indicate that genes indicative of aging may be successfully classified into DNA repair and non-DNA repair genes and such successful classification may help identify pathways and biomarkers that are important to the aging process.