GEASS: Neural causal feature selection for high-dimensional biological data

Authors: Mingze Dong, Yuval Kluger

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the efficacy of GEASS in several synthetic datasets and real biological data from single-cell RNA sequencing and spatial transcriptomics.
Researcher Affiliation Academia Mingze Dong Yale University mingze.dong@yale.edu Yuval Kluger Yale University yuval.kluger@yale.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the GEASS methodology described.
Open Datasets Yes The data preprocessing is consistent with the sc Velo tutorial: https://scvelo.readthedocs.io/ Velocity Basics/ (Bergen et al., 2020). The data is downloaded from Dryad and preprocessed with the standard Scanpy pipeline (Wolf et al., 2018).
Dataset Splits No We use the same training parameters in all time-series settings, with the key sparsity regularization parameter λ1 set with 0.04/0.05 based on a validation set, and the rest parameter settings are consistent with default. The paper mentions using a 'validation set' but does not specify the explicit split percentages, sample counts, or a citation to a predefined split for reproducibility.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The VAR-LINGAM algorithm is implemented in the Python package LINGAM...The PCMCI algorithm is implemented in the Python package Tigramite...The GVAR algorithm is implemented at https://github.com/i6092467/GVAR. The data is downloaded from Dryad and preprocessed with the standard Scanpy pipeline (Wolf et al., 2018). The data preprocessing is consistent with the sc Velo tutorial: https://scvelo.readthedocs.io/ Velocity Basics/ (Bergen et al., 2020). The paper mentions software packages but does not provide specific version numbers for them.
Experiment Setup Yes The parameter set: order=5, hidden_layer_size = 10, end_epoch=50, batch_size = 50, lmbd=1 is used throughout our study. We use the same training parameters in all time-series settings, with the key sparsity regularization parameter λ1 set with 0.04/0.05 based on a validation set, and the rest parameter settings are consistent with default. The parameter set: λ1 = 0.06, λ2 = 0.1.