A novel feature extraction approach for predicting human essential genes by XGBoost machine learning

Cancer remains a pressing issue in the medical field, with most cancer cells associated with genetic abnormalities. Therefore, the effective prediction of human essential genes has become one of the significant challenges in computational biology. Determining essential gene features is crucial in the process of predicting essential genes using computational methods. In this study, we proposed the $\mathrm{P}_{\mathrm{ij}} \mathrm{Z}$ Curve method, which is based on the Z Curve and integrates the relationships among different positions of codon nucleotides, to extract the features of essential genes. Then, we used the XGBGain-FR method to rank the gene features. Subsequently, the XGBoost computational method was used to construct models for identifying human essential genes under 5 -fold and 10 -fold cross-validation. The AUC values of our two models were 0.8462 and 0.8449, respectively. Additionally, the two models showed similar performance in terms of accuracy, specificity, and BACC. These results indicate that our proposed feature extraction method performed well in identifying human essential genes. Furthermore, the method demonstrated robustness under the XGBoost algorithm, underscoring its efficiency in feature extraction. Finally, we hope that the $\mathrm{P}_{\mathrm{ij}} \mathrm{Z}$ Curve can be effectively applied to other gene classification problems.