Participants must submit three outputs:

File: `prediction.h5ad`

An AnnData file containing predicted gene expression profiles normalized and log-transformed post-perturbation for 2,863 gene perturbations indicated in predict_perturbations.txt File is available in the data/ directory. .

Predictions should be stored in adata.X matrix with the corresponding perturbation identity recorded in adata.obs['gene'].

The set of genes (columns) included in the prediction is defined explicitly by genes_to_predict provided at inference time and the columns of adata.X must follow this order.

Note that the genes_to_predict list may change between validation and test phases, and your model must generate predictions for whichever set of genes is supplied. The maximum number of genes that could be included in genes_to_predict is 21,592 corresponding to the total number of genes in the dataset.

For each gene perturbation, we ask you to predict the gene expression profiles for 100 cells to quantify the distribution of each perturbation prediction. With N = len(genes_to_predict), the final prediction file is therefore required to have dimensions: [286,300 × N] (cells × genes_to_predict).

File: `predict_program_proportion.csv`

A CSV file reporting the predicted proportion of cells with enriched programs for each gene perturbation listed in predict_perturbations.txt.

The file should contain one row per perturbation with the following columns:

gene: should contain the perturbation name,
pre_adipo, adipo, lipo, and other: should specify the predicted proportion of cells in each corresponding state for that perturbation,
lipo_adipo: should be the ratio of lipo to adipo (representing the proportion of adipocytes with enriched lipogenic programs).

This file should thus have 2,863 rows and 6 columns Use .to_csv(index=False) when saving the file. . An example is available in the data/ directory.

File: `Method description.md`

We ask you to please write a small document outlining the approaches used to generate both the predictions and the estimated proportions of cells enriched for each program.

This should include sufficient details of the computational models employed and the procedures used to derive cell proportions.

The document should be organized into three sections, represented as titles in a Markdown file:

Method Description: Explain how your method works. (5-10 sentences)
Rationale: Describe the reasoning behind your model. (5-10 sentences)
Data and Resources Used: Specify the datasets and any other resources utilized. (5-10 sentences)

Notes:

A human will validate the content at the end of the competition.
Work deemed unsatisfactory may be disqualified.
This file must be provided during submission.
If content needs to be changed, you must re-submit with the new version.
The name must be Method Description.md; case does not matter.
Only non-empty and non-comment lines are considered.

Below is an example of how to format the file:


  # Method Description

  

  <!-- Explain how your method works. (5-10 sentences) -->

  Lorem ipsum dolor sit amet, consectetur adipiscing elit.

  Aliquam eget augue quis metus viverra vehicula sit amet lacinia odio.

  

  # Rationale

  

  <!-- Describe the reasoning behind your gene panel design. (5-10 sentences) -->

  Praesent dignissim ipsum vel leo eleifend, eget pulvinar mauris ornare.

  Duis efficitur lectus posuere iaculis dictum.

  

  # Data and Resources Used

  

  <!-- Specify the datasets and any other resources utilized. (5-10 sentences) -->

  Donec feugiat eros vel odio gravida venenatis.

  Nam et sem sit amet nisi vestibulum semper bibendum et libero.

If there's an obvious issue regarding the format, you'll receive an immediate notification.

Notebook users are required to use embedded files.