neurips2021_multimodal_topmethods

This repository is a collection of top methods submitted to the OpenProblems / NeurIPS 2021 competition for multimodal single-cell data integration (link).

Through the pipelines and code contained in this repository, you should be able to replicate the obtained scores for each of the top submissions.

List of methods

Task	Team	Method	Authors
joint_embedding	Guanlab-dengkw	Guanlab-dengkw	Kaiwen Deng
joint_embedding	Amateur	JAE	Qiao Liu , Wanwen Zeng , Chencheng Xu
joint_embedding	Living-Systems-Lab	LSL_AE	Sumeer Khan, Robert Lehman, Xabier Martinez De Morentin, Aidyn Ubingazhibov, Minxing Pang
match_modality	GLUE	CLUE	Zhi-Jie Cao , Xin-Ming Tu , Chen-Rui Xia
match_modality	Novel	Novel	Gleb Ryazantsev, Nikolay Russkikh, Igor I
predict_modality	Cajal	Cajal	Anna Laddach , Roman Laddach , Michael Shapiro
predict_modality	DANCE	DANCE	Hongzhi Wen, Jiayuan Ding, Wei Jin, Xiaoyan Li, Zhaoheng Li, Haoyu Han, Yuying Xie, Jiliang Tang
predict_modality	Guanlab-dengkw	Guanlab-dengkw	Kaiwen Deng
predict_modality	LS_lab	LS_Lab	Aidyn Ubingazhibov
predict_modality	Novel	Novel	Gleb Ryazantsev, Nikolay Russkikh, Igor I
predict_modality	scJoint	scJoint	Yu-Hsiu Chen , Sheng Wan, Tung-Yu Wu
predict_modality	AXX	SimpleMLP	Xueer Chen , Jiwei Liu

Dependencies

The dependencies of this repository are the same as those of the competition itself (link).

To run any of the methods, you need to download the required binaries and datasets and build the relevant docker containers first.

# download viash and nextflow
bin/init

# sync datasets to local
src/sync_datasets.sh

# build components and docker containers
bin/viash_build --max_threads 4

You can then rerun components by running the bash script located in the respective folders.

For example:

$ src/predict_modality/methods/cajal/test.sh

...

N E X T F L O W  ~  version 21.04.1
Pulling openproblems-bio/neurips2021_multimodal_viash ...
Launching `openproblems-bio/neurips2021_multimodal_viash` [trusting_woese] - revision: a28e0c22c5 [1.4.0]
executor >  local (5)
[4e/f4b91d] process > get_id_predictions (4)                                                            [100%] 4 of 4, cached: 3 ✔
[94/ffca9e] process > get_id_solutions (2)                                                              [100%] 4 of 4, cached: 4 ✔
[b3/92d144] process > bind_tsv_rows:bind_tsv_rows_process (meta_metric)                                 [100%] 1 of 1, cached: 1 ✔
[44/667c85] process > mse:mse_process (openproblems_bmmc_cite_phase2_PM_gex2adt)                        [100%] 4 of 4, cached: 3 ✔
[4f/8061cc] process > correlation:correlation_process (openproblems_bmmc_cite_phase2_PM_gex2adt)        [100%] 4 of 4, cached: 3 ✔
[ea/988a0f] process > check_format:check_format_process (openproblems_bmmc_cite_phase2_PM_gex2adt)      [100%] 4 of 4, cached: 3 ✔
[dc/393dca] process > final_scores:final_scores_process (output)                                        [100%] 1 of 1 ✔

[{"method_id":"cajal","ADT2GEX":0.3273,"ATAC2GEX":0.2172,"GEX2ADT":0.4613,"GEX2ATAC":0.178,"Overall":0.2959}]

After the bash script has finished running, output will have been stored in:

• output/pretrain/<task-id>/<method-id>/<dataset-id>.output_pretrain: Pre-trained models (if required).
• output/predictions/<task-id>/<method-id>/<dataset-id>: Predictions made by the method.
• output/evaluation<method-id>/<dataset-id>: Evaluation metrics.

Where is one of three competition tasks (predict_modality, match_modality or joint_embedding).