SimMISVM.jl

In this repository is the code associated with the publication titled "A Multi-Instance Support Vector Machine with Incomplete Data for Clinical Outcome Prediction of COVID-19" presented at the 12th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (BCB '21) held virtually from August 1-4, 2021.

This code base is using the Julia Language (v1.6.0) to make a reproducible scientific project named

SimMISVM.jl

To (locally) reproduce this project, do the following:

0. Download this code base.
1. Open a Julia console and do:

   julia> using Pkg
   julia> Pkg.activate("path/to/code")
   julia> Pkg.instantiate()

This will install all necessary packages for you to be able to run the scripts and everything should work out of the box.

2. Run the tests associated with the SimMISVM model:

   julia> include("test/simmisvm_test.jl")

These tests ensure that the updates derived in Algorithm 2 are correct. E.g. since variable update is derived with respect to a primal variable, and the minimization is quadratic with respect to that variable, the Lagrangian should be a minimum after that variable has been updated. Please note: The frist time this code is run it may take some extra time.

3. For an example on running the SimMISVM.jl model on the COVID-19 dataset run:

   julia> include("simmisvm_example.jl") # This is the main "entry-point"

4. The code for the updates in Algorithm 2 are located in src/SimMISVM.jl.

5. The hyperparameter settings for each method-dataset pair for the results reported in Table 1 are as follows:

Models implemented from: https://github.com/alan-turing-institute/MLJ.jl

Model	hyperparameter settings
kNN	K = 7
LightGBM	learning_rate = 0.27, num_leaves = 32, max_depth = 24
XGBoost	eta = 0.22, max_depth = 8, lambda = 0.73, alpha = 0.44
SVM	C = 5e4, kernel = linear

Our model:

Model	hyperparameter settings
MISVM	C = 1e-3, μ=1e-4, ρ=1.2
SimMISVM	C = 10, α=0.01, β=0.01, μ=1e-4, ρ=1.2

Other Files and Descriptions

• Manifest.toml and Project.toml specify the project's dependencies.
• data/raw_results.csv are used to generate Figure 2.
• data/time_series_375_prerpocess_en.csv is the raw COVID-19 patient data provided in https://www.nature.com/articles/s42256-020-0180-7.
• data_utils.jl contains functions that are used to assist in the handling of the temporal COVID-19 data. For integrating a new dataset please refer to this file.

Issues

If you have any trouble with this code please open a GitHub issue above.

Citing This Work

If you find this code useful please consider citing the following:

@inproceedings{brand2021multi,
  title={A multi-instance support vector machine with incomplete data for clinical outcome prediction of COVID-19},
  author={Brand, Lodewijk and Baker, Lauren Zoe and Wang, Hua},
  booktitle={Proceedings of the 12th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics},
  pages={1--6},
  year={2021}
}