Table 1.
Values of hyperparameters used in TabDEG model.
Fig 1.
Workflow of TabDEG.
Fig 2.
UMAP workflow.
Fig 3.
K-means clustering flowchart.
Fig 4.
The data is divided into two parts: disease group (T) and normal group (N). UMAP, KMEANS and PCA methods is used with aim to get mixed feature data in the process of DA. And then the data is standardized to ensure data scale consistency.
Fig 5.
(a) TabNet encoder, composed of a feature transformer, an attentive transformer and feature masking. A split block divides the processed representation and decomposition products obtained are used by the attentive transformer of the subsequent step as well as for the overall output. For each step, the feature selection mask provides interpretable information about the model’s functionality and the masks could be aggregated to obtain global feature important attribution. (b) TabNet decoder, composed of a feature transformer block at each step. (c) Attention transformer, the features of the n_a part enter the FC layer and expand to the same dimension as the input features to facilitate subsequent attention calculations, then enter the BN layer, and then multiply with “priors” before entering the sparsemax layer. The sparsemax layer is used for feature selection and “priors” is a constant vector with all entries being 1 in the initial step, which would change in each step. (d) Feature transformer, the output of the feature transformer is divided into two parts: n_d(n features for decision) and n_a(n features for attetion), where n_a is used for further calculation in subsequent steps and n_d is used for the final decision. (the output of the initial step has only n_a-dimensional features, rather than n_d-dimensional features to participate in the final decision).
Table 2.
Dataset abbreviations for cancer datasets.
Fig 6.
ROC curves with scores for different methods across ten datasets.
Table 3.
Performance of TabDEG against other five ML models on test data of all ten datasets with five-fold cross-validation being used.
Table 4.
Datasets GO ID/attribute p-value q-value.
Fig 7.
Pathways mapped from predicted UR and DR genes of BRCA.