Maximum n-times Coverage for Vaccine Design
Authors: Ge Liu, Alexander Dimitrakakis, Brandon Carter, David Gifford
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically evaluate the NTIMES-ILP and MARGINALGREEDY algorithms with two toy examples: a LARGE dataset where a set of 30 overlays are randomly generated to cover a set of 10 elements (with equal weights) 0, 1, or 2 times, and a SMALL dataset where a set of 10 overlays that were randomly generated to cover a set of 5 elements with equal weights 0, 1, or 2 times. Figure 2 shows the efficiency of the NTIMES-ILP and MARGINALGREEDY algorithms on both the n-TIMES SET COVER and MAXIMUM n-TIMES COVERAGE problems for varying values of n. |
| Researcher Affiliation | Academia | Ge Liu, Alexander Dimitrakakis, Brandon Carter, David Gifford Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology gifford@mit.edu |
| Pseudocode | Yes | Algorithm 1 MARGINALGREEDY algorithm (for MIN-COST n-TIMES COVERAGE) |
| Open Source Code | Yes | We provide an open-source implementation of our methods at https://github.com/gifford-lab/optivax. |
| Open Datasets | Yes | Models and haplotype frequencies (Liu et al., 2020) were obtained from Mendeley Data (https://doi.org/10.17632/cfxkfy9zp4.1, https://doi.org/10.17632/gs8c2jpvdn.1) and are available under a Creative Commons Attribution 4.0 International license. |
| Dataset Splits | No | The paper does not provide specific training/validation/test dataset splits for its own experimental setup or model training. It mentions 'validation' in the context of evaluation of existing ML models or the validation set in general, but not for its own data partitioning. |
| Hardware Specification | Yes | We utilized Google Cloud Platform machines with 224 CPU cores for Marginal Greedy optimization and parallelized computation across all CPU cores. For ILP designs, we used our own computing resources with 8 CPU cores. The prediction of peptide-HLA binding with machine learning models (Net MHCpan, Net MHCIIpan, MHCflurry, PUFFIN) was done using our own computing resources with 200 CPU cores and NVIDIA Ge Force RTX 2080 Ti GPUs. |
| Software Dependencies | No | The paper mentions 'Python-MIP solver' but does not specify a version number. It also mentions 'SCIP (Achterberg, 2009)' and 'LMHS (Saikko et al., 2016)' but these are cited papers, not the versions of the software used by the authors. |
| Experiment Setup | Yes | We directly applied MARGINALGREEDY on Eval Vax-Robust objective function with a beam size of 10 for MHC class I and 5 for MHC class II, but we further reduced peptide redundancy by eliminating unselected peptides that are within three (MHC class I) or five (MHC class II) edits on a sequence distance metric from the selected peptides at each iteration. |