Search for a command to run...
In our model Sparse Modality KAN, we have used the Tox 21 dataset for chemical toxicity prediction where each row corresponds to a chemical compound, and each column represents a specific molecular feature or fragment.Since any single molecule contains only a small subset of all possible chemical substructures, the majority of feature values are zero, leading to high sparsity. The Fourier fusion method based on the KAN approach facilitated the interaction of different feature space dimensions, making it possible to learn the complex relationship using the model. This approach enhanced the predictive capability of the model and showed greater generalization abilities than other machine learning techniques using simple architectures In summary, the experimental outcomes show the effectiveness of the proposed approach for distinguishing toxic from non-toxic molecules using the proposed mode.