Bayesian Causal Structural Learning with Zero-Inflated Poisson Bayesian Networks

Authors: Junsouk Choi, Robert Chapkin, Yang Ni

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the utility of the proposed ZIPBN in causal discoveries for zero-inflated count data by simulation studies with comparison to alternative Bayesian network methods. Additionally, real single-cell RNA-sequencing data with known causal relationships will be used to assess the capability of ZIPBN for discovering causal relationships in real-world problems.
Researcher Affiliation Academia Junsouk Choi Department of Statistics Texas A&M University College Station, TX 77843 jchoi@stat.tamu.edu Robert Chapkin Department of Nutrition Texas A&M University College Station, TX 77843 r-chapkin@tamu.edu Yang Ni Department of Statistics Texas A&M University College Station, TX 77843 yni@stat.tamu.edu
Pseudocode Yes Algorithm 1 Parallel-Tempered MCMC for ZIPBN
Open Source Code No The paper states 'The code implementing the MCMC is available in the Supplementary Material.' This does not explicitly state it is open-source or provide a direct public link, which is required by the prompt's criteria.
Open Datasets Yes Additionally, real single-cell RNA-sequencing data with known causal relationships will be used to assess the capability of ZIPBN for discovering causal relationships in real-world problems.
Dataset Splits No The paper mentions 'simulated data under different samples sizes n {250, 500, 1000}' and 'retained 479 pairs for causal validation', but does not provide explicit train/validation/test splits with percentages or sample counts for reproducibility.
Hardware Specification Yes The CPU time was 1.7 hours on an i9-9880H 2.3GHz CPU.
Software Dependencies No The paper mentions 'R package Seurat (Butler et al., 2018)' but does not provide specific version numbers for R, Seurat, or any other software components.
Experiment Setup Yes For the proposed ZIPBN, we used non-informative prior by setting the hyperparameters to be (aτ, bτ) = (0.01, 0.01) and (aρ, bρ) = (0.5, 0.5)... We ran M = 10 parallel chains for 3, 000 iterations, of which the first 1, 500 iterations were discarded as burn-in. The temperatures were chosen uniformly between 0 and 1 on the log-scale, i.e., log(Tm) = (m 1)/9 for m = 1, . . . , 10. The swapping probability ps was chosen to be 10%.