Pathway Graphical Lasso
Authors: Maxim Grechkin, Maryam Fazel, Daniela Witten, Su-In Lee
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compared Path GLasso with QUIC, HUGE and DPGLASSO on 3 scenarios: 1) Cycle: Pathways form one large cycle with 50 genes per pathway with overlap size of 10; 2) Lattice: The true underlying model is a 2D lattice, and each pathway contains between 3 and 7 nearby variables; and 3) Random: Each pathway consists of randomly selected genes. For each setting, we generated a true underlying connectivity graph, converted it to the precision matrix following the procedure from (Liu and Ihler 2011), and generated 100 samples from the multivariate Gaussian distribution. We observed that Path GLasso dramatically improves the run time compared to QUIC, HUGE and DP-GLASSO (Figure 4), sometimes up to two orders of magnitude. |
| Researcher Affiliation | Academia | Maxim Grechkin University of Washington Seattle, WA grechkin@uw.edu Maryam Fazel University of Washington Seattle, WA mfazel@uw.edu Daniela Witten University of Washington Seattle, WA dwitten@uw.edu Su-In Lee University of Washington Seattle, WA suinlee@uw.edu |
| Pseudocode | No | The paper describes algorithms using text and mathematical expressions, but does not contain structured pseudocode or algorithm blocks (clearly labeled algorithm sections or code-like formatted procedures). |
| Open Source Code | Yes | The implementation of Path GLasso and example data are available at: http://pathglasso-leelab.cs.washington.edu/. |
| Open Datasets | Yes | We considered two gene expression datasets from acute myeloid leukemia (AML) studies: MILE (Haferlach et al. 2010) and GENTLES (Gentles et al. 2010) containing 541 and 248 samples, respectively. |
| Dataset Splits | No | The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) for training, validation, and test sets. It mentions training on MILE data and testing on GENTLES data but no internal splits or validation set details. |
| Hardware Specification | Yes | All comparisons were run on 4 core Intel Core i7-3770 CPU @ 3.40GHz with 8GB of RAM. |
| Software Dependencies | No | The paper mentions software like 'R package HUGE' and 'DP-GLASSO' but does not provide specific version numbers for these or any other software dependencies, which are needed to replicate the experiment. |
| Experiment Setup | Yes | We set λ = 1010 for the entries that lie outside of the pathways, making them solve exactly the same problem as Path GLasso. |