How-to: Train InstaNovo Models
This guide explains how to train your own InstaNovo and InstaNovo+ models.
Preparing your data
Before you can start training, you need to have your data in a format that InstaNovo can understand. The easiest way to do this is to use the SpectrumDataFrame. See the how-to guide on using custom datasets for more details.
Training InstaNovo (Transformer)
The InstaNovo transformer model is the base model that performs the initial de novo sequencing.
To train the transformer model, you use the instanovo transformer train command. The training process is configured using a YAML file. The default configuration file is instanovo/configs/instanovo.yaml.
To start training with the default configuration, you would run:
You will likely want to customize the training configuration, such as the paths to your training and validation data. You can do this by creating your own YAML file and passing it to the command, or by overriding specific values from the command line.
For example, to specify the training and validation data paths, you could run:
instanovo transformer train --data.train_data_path /path/to/train/data --data.val_data_path /path/to/val/data
To see all the available options for the training command, run:
To customize the model architecture, you can modify the model configuration file at instanovo/configs/model/instanovo_base.yaml.
Training InstaNovo+ (Diffusion)
The InstaNovo+ diffusion model is used to refine the predictions made by the transformer model.
Training the diffusion model is similar to training the transformer model. You use the instanovo diffusion train command, and the configuration is managed through a YAML file (default: instanovo/configs/instanovoplus.yaml).
To start training the diffusion model, you would run:
Again, you can customize the configuration by providing your own YAML file or by overriding values from the command line.
To see all the available options, run:
To customize the model architecture, you can modify the model configuration file at instanovo/configs/model/instanovoplus_base.yaml.