Fine-tuning for batch integration
The model card indicates that fine-tuning can be done for batch integration, but I cannot find any examples/tutorials/detailed guides on how this was done in the model card or manuscript. Are there any resources available for this? Ideally I would appreciate something like this (https://github.com/bowang-lab/scGPT/blob/main/tutorials/Tutorial_Integration.ipynb) -- scGPT's tutorial for fine-tuning for batch correction in Jupyter notebook form.
Thank you for your question. When you fine-tune Geneformer to a task, for example disease state classification, the batches will be integrated by virtue of the model focusing in on the pertinent features for that distinction, as opposed to the features that distinguish batches. So, the recommendation would be to fine-tune towards what is most important for answering your biological question, not for the sole purpose of integrating batches, which would be a preprocessing task but not addressing the biological question at hand.