fit_model {ttgsea}

R Documentation

Deep learning model fitting

Description

From the result of GSEA, we can predict enrichment scores for unique tokens or words from text in names of gene sets by using deep learning. The function "text_token" is used for tokenizing text and the function "token_vector" is used for encoding. Then the encoded sequence is fed to the embedding layer of the model.

Usage

fit_model(gseaRes, text, score, model, ngram_min = 1, ngram_max = 2,
          num_tokens, length_seq, epochs, batch_size,
          use_generator = TRUE, ...)

Arguments

gseaRes	a table with GSEA result having rows for gene sets and columns for text and scores
text	column name for text data
score	column name for enrichment score
model	deep learning model, input dimension and length for the embedding layer must be same to the "num_token" and "length_seq", respectively
ngram_min	minimum size of an n-gram (default: 1)
ngram_max	maximum size of an n-gram (default: 2)
num_tokens	maximum number of tokens, it must be equal to the input dimension of "layer_embedding" in the "model"
length_seq	length of input sequences, it must be equal to the input length of "layer_embedding" in the "model"
epochs	number of epochs
batch_size	batch size
use_generator	if "use_generator" is TRUE, the function "sampling_generator" is used for "fit_generator". Otherwise, the "fit" is used without a generator.
...	additional parameters for the "fit" or "fit_generator"

Value

model	trained model
tokens	information for tokens
token_pred	prediction for every token, each row has a token and its predicted score
token_gsea	list of the GSEA result only for the corresponding token
num_tokens	maximum number of tokens
length_seq	length of input sequences

Author(s)

Dongmin Jung

See Also

keras::fit_generator, keras::layer_embedding, keras::pad_sequences, textstem::lemmatize_strings, text2vec::create_vocabulary, text2vec::prune_vocabulary

Examples

library(reticulate)
if (keras::is_keras_available() & reticulate::py_available()) {
  library(fgsea)
  data(examplePathways)
  data(exampleRanks)
  names(examplePathways) <- gsub("_", " ",
                            substr(names(examplePathways), 9, 1000))
  set.seed(1)
  fgseaRes <- fgsea(examplePathways, exampleRanks)
  
  num_tokens <- 1000
  length_seq <- 30
  batch_size <- 32
  embedding_dims <- 50
  num_units <- 32
  epochs <- 1
  
  ttgseaRes <- fit_model(fgseaRes, "pathway", "NES",
                         model = bi_gru(num_tokens,
                                        embedding_dims,
                                        length_seq,
                                        num_units),
                         num_tokens = num_tokens,
                         length_seq = length_seq,
                         epochs = epochs,
                         batch_size = batch_size,
                         use_generator = FALSE)
}

---