Here we report some fdsa usage examples.
Example data can be downloaded here.
The set matching network simply plays the role of a mapper in a set-autoencoder and is not optimised during training. Therefore, it must first be pre-trained before it can be plugged into the set autoencoder.
The synthetic data to train the set matching network can be generated during runtime
using the pytoda package. The dataset can be customised with respect to its
distribution, shape, and the cost metric used to generate the assignment targets for
supervised learning. See the training scripts in set_matching folder
on how to create and use these custom synthetic datasets.
The example rnn_setmatching.py performs a comparative analysis of RNNs using 4 recurrent cells, namely, GRU, LSTM, nBRC and BRC in the set matching task. The parameters for training are provided as JSON files.
(fdsa) $ python examples/set_matching/rnn/rnn_setmatching.py -h
usage: rnn_setmatching.py [-h] model_path results_path training_params
positional arguments:
model_path Path to save best model.
results_path Path to save results and plots.
training_params Path to the training parameters.
optional arguments:
-h, --help show this help message and exitThe RNN models return a set2 vs set1 constrained similarity matrix, and so the predictions
are compared with target21.
The example seq2seq_setmatching.py performs a comparative analysis of Sequence2Sequence models with attention using 4 recurrent cells, namely, GRU, LSTM, nBRC and BRC as the encoding/decoding unit in the set matching task. The parameters for training are provided as JSON files.
(fdsa) $ python examples/set_matching/seq2seq/seq2seq_setmatching.py -h
usage: seq2seq_setmatching.py [-h] model_path results_path training_params
positional arguments:
model_path Path to save best model.
results_path Path to save results and plots.
training_params Path to the training parameters.
optional arguments:
-h, --help show this help message and exitThe Seq2Seq models return a set1 vs set2 constrained similarity matrix, and so the
predictions are compared with target12.
NOTE: The dimensions along which the argmax is computed in the constrained similarity matrix
decides whether target12 or target21 are used as the true targets for loss calculation.
target12 are the row-wise non-zero indices in a set1 vs set2 constrained similarity matrix,
and target21 are the row-wise non-zero indices in a set2 vs set1 constrained similarity matrix.
The set_autoencoder folder contains 2 examples for two tasks - reconstructing shapes and reconstructing 128D cancer data.
This tasks serves as a sanity check for the set autoencoder and so only uses the Hungarian algorithm.
The data for the shapes task consists of randomly generated 2D point clouds of squares, crosses, and circles,
of various sizes and positions, and made up of a varying number of points. The data can be generated
using the shapes_data.py script.
The data are saved as a .csv file, which is then passed into the shapes_train.py
script. This script performs a comparative analysis of using GRU, LSTM, pLSTM and nBRC recurrent cells
as the encoding/decoding unit in the set autoencoder on reconstructing 2D shapes. The
training parameters are provided as a JSON file.
(fdsa) $ python examples/set_autoencoder/shapes/shapes_train.py -h
usage: shapes_train.py [-h]
model_path results_path training_data_path
validation_data_path testing_data_path training_params
positional arguments:
model_path Path to save the best performing model.
results_path Path to save the training and validation losses.
training_data_path Path to the training data.
validation_data_path Path to the validation data.
testing_data_path Path to the testing data.
training_params Path to the training parameters json file.
optional arguments:
-h, --help show this help message and exitThe data are provided as torch files that contain tensors of transcriptomic and protein data
combined into sets, and split into training, validation and test datasets, such that no cell line
or protein seen during training are validated or tested. Additionally, the permutations
used in shuffling the order of the elements of the set are also provided.
The files can be found here.
The training script setsae_setm_train.py generates reconstructions for the specified parameters, which are provided as JSON files. The Hungarian or Gale-Shapley algorithm can be used in place of the pre-trained network as a baseline by commenting and uncommenting the lines pertaining to the mapper. Note: Even if the experiment is for the Hungarian/Gale-Shapley algorithm, a dummy file for matching parameters and model path should be provided to maintain the flexibility of switching between the network and algorithms.
The pre-trained matching network can be found here.
(fdsa) $ python examples/set_autoencoder/reconstruct_128D/setsae_setm_train.py
usage: setsae_setm_train.py [-h]
model_path results_path training_params
matching_params train_data_path valid_data_path
test_data_path
positional arguments:
model_path Path where the pre-trained matching network is saved.
results_path Path to save the results, logs and best model.
training_params Path to the training parameters json file.
matching_params Path to the matching network parameters json file.
train_data_path Path to the training data.
valid_data_path Path to the validation data.
test_data_path Path to the testing data.
optional arguments:
-h, --help show this help message and exitThe example mol_gen.py uses the encoder of the pre-trained set autoencoder to produce embeddings of sets of transcriptomic and proteomic data. This embedding acts as a multi-modal context for molecule generation in the Paccmann^{RL} generative model to generate candidate drugs against a given cancer type.
The experiment makes use of sets of transcriptomic profiles of cell lines and associated proteins to condition molecule generation. A LOOCV on the cell-lines is used to evaluate the performance of this model, that is, the test set comprises a cartesian product of the test cell-line and all proteins under consideration.
The molecule_generation folder contains the necessary JSON parameter files for training. The omics data, protein data and unbiased predictions can be found here here. The pre-trained models required to succesfully run the script can be found as follows:
- Encoder model (encoder_model_path, encoder_params_path) and parameters can be found here
- Molecule model (mol_model_path) can be found here
- IC50 model (ic50_model_path) can be found here
- Affinity model (affinity_model_path) can be found here
- Tox21 model (tox21_path) can be found here under pretraining/toxicity_predictor.
(fdsa) $ python examples/molecule_generation/mol_gen.py -h
usage: mol_gen.py [-h] [--test_protein_name TEST_PROTEIN_NAME]
[--tox21_path TOX21_PATH]
omics_data_path protein_data_path test_cell_line
encoder_model_path mol_model_path ic50_model_path
affinity_model_path params_path encoder_params_path
results_path unbiased_protein_path unbiased_omics_path site
model_name
PaccMann^RL training script
positional arguments:
omics_data_path Omics data path to condition molecule generation.
protein_data_path Protein data path to condition molecule generation.
test_cell_line Name of testing cell line (LOOCV).
encoder_model_path Path to setAE model.
mol_model_path Path to chemistry model.
ic50_model_path Path to pretrained IC50 model.
affinity_model_path Path to pretrained affinity model.
params_path Directory containing the model params JSON file.
encoder_params_path directory containing the encoder parameters JSON file.
results_path Path where results are saved.
unbiased_protein_path
Path where unbiased protein predictions are saved.
unbiased_omics_path Path where unbiased omics predictions are saved.
site Name of the cancer site for conditioning generation.
model_name Name for the trained model.
optional arguments:
-h, --help show this help message and exit
--test_protein_name TEST_PROTEIN_NAME
Optional gene name of testing protein (LOOCV).
--tox21_path TOX21_PATH
Optional path to Tox21 model.For more examples see other repositories in the PaccMann organization.