Closed YOHOHO111 closed 3 months ago
Also, i tried the sliding window to cut the sequence, but got too much result, and i haven't find a way to deal with them, and it also cannot used to train the data directly.
try using v,j segm and mhc as feature, using kmer to encode the cdr3 sequence, combine all the feature to a matrix, get a better result with the new matrix
simplify the model structure by replacing RNN with DNN
The result is all not ideal, the accuracy of mouse combined chain is about 0.60, the loss is about 1, but better in val_acc and val_loss, the accuracy is even worse in human combined chain, which is about 0.30 and 2 in loss, and val_acc and val_loss is even worse.