SHAP zero

These files contain data from the article "SHAP zero Explains All-order Feature Interactions in Black-box Genomic Models with Near-zero Query Cost" by Darin Tsui, Aryan Musharaf, and Amirali Aghazadeh. Files presented here contains all SHAP values and Shapley interactions found by SHAP zero, KernelSHAP, and SHAP-IQ from inDelphi and TIGER, using code from https://github.com/amirgroup-codes/shapzero/.Description of filesTIGER_shapzero_values.csv: All SHAP values found by SHAP zero from TIGER.Columns:Sequence: sequence inputted to TIGER.Position 1 to 26: SHAP value of each nucleotide at each position.TIGER_kernelshap_values.csv: All SHAP values found by KernelSHAP from TIGER.Columns:Sequence: sequence inputted to TIGER.Position 1 to 26: SHAP value of each nucleotide at each position.TIGER_deepshap_values.csv: All SHAP values found by DeepSHAP from TIGER.Columns:Sequence: sequence inputted to TIGER.Position 1 to 26: SHAP value of each nucleotide at each position.TIGER_fastshap_values.csv: All SHAP values found by FastSHAP from TIGER.Columns:Sequence: sequence inputted to TIGER.Position 1 to 26: SHAP value of each nucleotide at each position.TIGER_shapzero_fsi.csv: All Faith-Shap interactions found by SHAP zero from TIGER.Columns:Sequence: sequence inputted to TIGER.Order 1 to 5: Dictionary of all interactions of orders 1-5. The keys represent the nucleotide position (0-indexed) affected, and the values represent the Faith-Shap interaction at those nucleotides. For example, the key:value pairing {(0,1,2):0.1} and the sequence TCA, means that the first (T), second (C), and third (A) nucleotides are interacting with a Faith-Shap interaction value of 0.1.TIGER_shapiq_fsi.csv: All Faith-Shap interactions found by SHAP-IQ from TIGER.Columns:Sequence: sequence inputted to TIGER.Order 1 to 2: Dictionary of all interactions of orders 1-2. The keys represent the nucleotide position (0-indexed) affected, and the values represent the Faith-Shap interaction at those nucleotides.Order 3 part 1 to 6: Since SHAP-IQ produced too many 3rd-order interactions for Excel to handle in a single column, the 3rd orders are split across 6 columns.inDelphi_shapzero_values.csv: All SHAP values found by SHAP zero from inDelphi.Columns:Sequence: sequence inputted to inDelphi.Position 1 to 40: SHAP value of each nucleotide at each position.inDelphi_kernelshap_values.csv: All SHAP values found by KernelSHAP from inDelphi.Columns:Sequence: sequence inputted to inDelphi.Position 1 to 40: SHAP value of each nucleotide at each position.inDelphi_fastshap_values.csv: All SHAP values found by FastSHAP from inDelphi.Columns:Sequence: sequence inputted to inDelphi.Position 1 to 40: SHAP value of each nucleotide at each position.inDelphi_shapzero_fsi.csv: All Faith-Shap interactions found by SHAP zero from inDelphi.Columns:Sequence: sequence inputted to inDelphi.Order 1 to 5: Dictionary of all interactions of orders 1-5. The keys represent the nucleotide position (0-indexed) affected, and the values represent the Faith-Shap interaction at those nucleotides. For example, the key:value pairing {(0,1,2):0.1} and the sequence TCA, means that the first (T), second (C), and third (A) nucleotides are interacting with a Faith-Shap interaction value of 0.1.inDelphi_shapiq_fsi.csv: All Faith-Shap interactions found by SHAP-IQ from inDelphi.Columns:Sequence: sequence inputted to inDelphi.Order 1 to 2: Dictionary of all interactions of orders 1-2. The keys represent the nucleotide position (0-indexed) affected, and the values represent the Faith-Shap interaction at those nucleotides.Order 3 part 1 to 20: Since SHAP-IQ produced too many 3rd-order interactions for Excel to handle in a single column, the 3rd orders are split across 20 columns.