Collision cross-collision section prediction
============================================

3DMolMS can be used to predict MS/MS-related properties, such as retention time (RT) and collision cross section (CCS). This guide shows how to train a model for RT prediction and CCS prediction, and how to transfer these models to your own RT and CCS dataset.

All models mentioned can be downloaded from `release v1.2.0 <https://github.com/JosieHong/3DMolMS/releases/tag/v1.2.0>`_.

Setup
-----

Please set up the environment as shown in the :doc:`../sourcecode` page.

**Step 1**: Data preparation
----------------------------

Download the cross-collision section dataset, `AllCCS <http://allccs.zhulab.cn/>`_, manually or using ``download_allccs.py``:

.. code-block:: bash

   python scripts/download_allccs.py --user <user_name> --passw <passwords> --output ./data/origin/allccs_download.csv

The structure of data directory is:

.. code-block:: text

   |- data
     |- origin
       |- allccs_download.csv

**Step 2**: Preprocessing
-------------------------

Use the following commands to preprocess the datasets. The settings of datasets are in ``./molnetpack/config/preprocess_etkdgv3.yml``.

.. code-block:: bash

   python scripts/preprocess.py --task ccs \
   --data_config_path ./molnetpack/config/preprocess_etkdgv3.yml

**Step 3**: Training
--------------------

Use the following commands to train the model. The settings of model and training are in ``./molnetpack/config/molnet_ccs.yml``.

*Using the command-line script:*

Learning from scratch:

.. code-block:: bash

   python scripts/train.py --task ccs \
   --train_data ./data/allccs_etkdgv3_train.pkl \
   --test_data ./data/allccs_etkdgv3_test.pkl \
   --checkpoint_path ./check_point/molnet_ccs_etkdgv3.pt

If you'd like to train this model from the pre-trained model on MS/MS prediction, please download the pre-trained model from `release v1.2.0 <https://github.com/JosieHong/3DMolMS/releases/tag/v1.2.0>`_.

Learning from pretrained model:

.. code-block:: bash

   python scripts/train.py --task ccs \
   --train_data ./data/allccs_etkdgv3_train.pkl \
   --test_data ./data/allccs_etkdgv3_test.pkl \
   --checkpoint_path ./check_point/molnet_ccs_etkdgv3_tl.pt \
   --transfer \
   --resume_path ./check_point/molnet_qtof_etkdgv3.pt

*Using the Python API:*

.. code-block:: python

   import torch
   from molnetpack import MolNet

   device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
   molnet_engine = MolNet(device, seed=42)

   # Learning from scratch:
   molnet_engine.train(
       task='ccs',
       train_data='./data/allccs_etkdgv3_train.pkl',
       valid_data='./data/allccs_etkdgv3_test.pkl',
       checkpoint_path='./check_point/molnet_ccs_etkdgv3.pt',
   )

   # Learning from pretrained MS/MS model (transfer learning):
   molnet_engine.train(
       task='ccs',
       train_data='./data/allccs_etkdgv3_train.pkl',
       valid_data='./data/allccs_etkdgv3_test.pkl',
       checkpoint_path='./check_point/molnet_ccs_etkdgv3_tl.pt',
       resume_path='./check_point/molnet_qtof_etkdgv3.pt',
       transfer=True,
   )