38512d92a8682a62e73c5b9e86366888be374532,dragonn/interpret/ism.py,,in_silico_mutagenesis,#,15

Before Change

    functor=get_logit_function(model,target_layer_idx)
    /Ǘ. get the wildtype predictions (n,1)                                                                                                                                                         
    wild_type_logits=np.asarray(get_logit(functor,X))
    wild_type_logits=np.squeeze(wild_type_logits,axis=2)
    /ǘ. expand the wt array to dimensions: (num_samples,1,sequence_length)                                                                                                                         
    //Initialize mutants array to the same shape                                                                                                                                                              
    output_dim=wild_type_logits.shape+X.shape[2:4]
    wt_expanded=np.empty(output_dim)
    mutants_expanded=np.empty(output_dim)
    empty_onehot=np.zeros(output_dim[3])
    /Ǚ. Iterate through all tasks, positions                                                                                                                                                                 
    for task_index in range(output_dim[0]):
        for sample_index in range(output_dim[1]):
            print("task:"+str(task_index)+" sample:"+str(sample_index))
            //fill in wild type logit values into an array of dim (task,sequence_length,num_bases)                                                                                                                    
            wt_logit_for_task_sample=wild_type_logits[task_index][sample_index]
            wt_expanded[task_index][sample_index]=np.tile(wt_logit_for_task_sample,(output_dim[2],output_dim[3]))
            //mutagenize each position                                                                                                                                                                            
            for base_pos in range(output_dim[2]):
                //for each position, iterate through the 4 bases                                                                                                                                                  
                for base_letter in range(output_dim[3]):
                    cur_base=np.copy(empty_onehot)
                    cur_base[base_letter]=1
                    Xtmp=np.copy(np.expand_dims(X[sample_index],axis=0))
                    Xtmp[0][0][base_pos]=cur_base
                    //get the logit of Xtmp                                                                                                                                                                              
                    mutants_expanded[task_index][sample_index][base_pos][base_letter]=np.squeeze(get_logit(functor,Xtmp)[task_index])
    //subtract wt_expanded from mutants_expanded
    ism_vals=mutants_expanded-wt_expanded
    //For each position subtract the mean ISM score for that position from each of the 4 values
    ism_vals_mean=np.expand_dims(np.mean(ism_vals,axis=3),axis=3)
    ism_vals_normed=ism_vals-ism_vals_mean

After Change

                Xtmp=np.array(np.expand_dims(X[sample_index],axis=0))
                Xtmp[0][0][base_pos]=cur_base
                //get the logit of Xtmp
                Xtmp_logit=np.squeeze(preact_function(Xtmp),axis=0)
                mutants_expanded[sample_index][0][base_pos][base_letter]=Xtmp_logit[task_index]
    //subtract wt_expanded from mutants_expanded
    ism_vals=mutants_expanded-wt_expanded
    //For each position subtract the mean ISM score for that position from each of the 4 values

Instances