KMOU-A-mobility-informed-SIR-model

Research code and reproducibility materials for the manuscript:

A Mobility-Informed SIR Model for Evaluating the Effects of Mobility Reduction on Epidemic Transmission and the instantaneous Reproduction Number

Authors: Jaewan Shin and Minsoo Kim
Affiliation: Major of Data Science, Korea Maritime & Ocean University, Busan, Republic of Korea

This repository contains the notebooks and supporting documentation used to construct subway-ridership-based mobility coefficients, estimate time-varying transmission dynamics with a mobility-informed SIR model, and evaluate mobility-reduction scenarios for influenza transmission.

Overview

This study proposes a mobility-informed SIR model for epidemic simulation when detailed origin-destination mobility matrices are unavailable.

The workflow uses daily aggregated subway boarding counts to construct a time-varying mobility coefficient, denoted by (\xi(t)) in the manuscript. The coefficient modifies the infection transmission term in the SIR model. Time-varying transmission rates (\beta(t)) and instantaneous reproduction numbers (R(t)) are then estimated using particle smoothing under a negative binomial observation model.

The repository is organized around the five uploaded notebooks:

Order	Notebook	Main role
1	`notebooks/01_mobility_factor_2016_2017.ipynb`	Constructs (m(t)), (L), (c(t)), (b(t)), and mobility-reduction scenario columns from subway boarding data
2	`notebooks/02_sir_negative_binomial_particle_smoothing.ipynb`	Validates particle filtering and particle smoothing using a synthetic SIR trajectory with a negative binomial observation model
3	`notebooks/03_rt_estimation_2016_2017.ipynb`	Estimates (\beta(t)), simulated incidence, and (R(t)) for the 2016-2017 influenza season
4	`notebooks/04_rt_heatmap_scenarios.ipynb`	Generates heatmaps of (R(t)) under observed mobility and 10-50% mobility-reduction scenarios
5	`notebooks/05_scenario_summary_tables.ipynb`	Generates scenario summary metrics such as mean (R(t)), days with (R(t)>1), and excess transmission area

Important source-code notation

The manuscript denotes the mobility coefficient as:

[ \xi(t) ]

However, the uploaded mobility-factor notebook stores this same mobility coefficient in columns named:

gamma
gamma_1
gamma_2
gamma_3
gamma_4
gamma_5

In this repository, these column names are preserved because they are used by the attached notebooks. They should be interpreted as follows:

Code column	Manuscript notation	Meaning
`gamma`	(\xi(t))	Observed subway-ridership-based mobility coefficient
`gamma_1`	(0.9\xi(t))	10% mobility-reduction scenario
`gamma_2`	(0.8\xi(t))	20% mobility-reduction scenario
`gamma_3`	(0.7\xi(t))	30% mobility-reduction scenario
`gamma_4`	(0.6\xi(t))	40% mobility-reduction scenario
`gamma_5`	(0.5\xi(t))	50% mobility-reduction scenario

The SIR recovery rate is not stored in these mobility-factor columns. In the Rt(2016~2017).ipynb notebook, the recovery rate is represented by:

sigma = 1.0 / 4.1

No theta parameter is used in the attached manuscript or the attached notebooks.

Study scope

The manuscript analyzes four influenza seasons:

Season label	Analysis period
`2016~2017`	2016-09-01 to 2017-08-31
`2017~2018`	2017-09-01 to 2018-08-31
`2018~2019`	2018-09-01 to 2019-08-31
`2022~2023`	2022-09-01 to 2023-08-31

The model is applied independently to five analysis labels:

Seoul
Busan
Daegu
Daejeon
Gwangju

The attached executable Rt(2016~2017).ipynb notebook focuses on the 2016-2017 season. The heatmap and scenario-table notebooks expect (R(t)) files for all four seasons.

Model summary

The mobility-informed SIR model is:

[ \frac{dS}{dt} = -\xi(t)\beta(t)\frac{S(t)I(t)}{N}, ]

[ \frac{dI}{dt} = \xi(t)\beta(t)\frac{S(t)I(t)}{N} - \gamma I(t), ]

[ \frac{dR}{dt} = \gamma I(t). ]

The mobility-weighted instantaneous reproduction number is:

[ R(t)=\frac{\xi(t)\beta(t)}{\gamma}\frac{S(t)}{N}. ]

The recovery parameter is fixed using a mean infectious period of 4.1 days:

[ \gamma^{-1}=4.1. ]

Mobility-factor construction

The mobility coefficient is constructed from daily subway boarding counts.

Let (X(t)) be the total number of subway boardings on day (t), and let (Pop) be the population denominator. The per-capita mobility intensity is:

[ m(t)=\frac{X(t)}{Pop}. ]

The baseline mobility level in the manuscript is:

[ L=\frac{1}{|T|}\sum_{t\in T}m(t). ]

The normalized mobility variation is:

[ c(t)=\frac{m(t)}{L}. ]

A centered 7-day moving average is applied:

[ b(t)=\frac{1}{|W(t)|}\sum_{\mu\in W(t)}c(\mu), \quad W(t)={\mu\in T: |\mu-t|\leq 3}. ]

The final manuscript mobility factor is:

[ \xi(t)=b(t). ]

Implementation note: the uploaded mobility factor(2016~2017).ipynb notebook first creates m_7, a centered 7-day moving average of m, and then computes L as the city-level mean of m_7. The notebook then computes c = m / L, b as a centered 7-day moving average of c, and writes gamma = b. This implementation note is documented here so that the GitHub explanation follows the attached code exactly.

Mobility-reduction scenarios

The uploaded mobility-factor notebook creates five reduced-mobility columns:

[ 0.9\xi(t),;0.8\xi(t),;0.7\xi(t),;0.6\xi(t),;0.5\xi(t). ]

These correspond to 10%, 20%, 30%, 40%, and 50% reductions in observed mobility.

For each scenario, the notebook computes:

[ R_c(t)=\frac{\xi_c(t)\beta(t)}{\gamma}\frac{S(t)}{N}. ]

The scenario summary notebook computes:

[ \Delta R_c(t)=R_0(t)-R_c(t), ]

[ D_c=\sum_t \mathbf{1}{R_c(t)>1}, ]

[ E_c=\sum_t \max(R_c(t)-1,0). ]

Repository structure

KMOU-A-mobility-informed-SIR-model/
├── README.md
├── CITATION.cff
├── LICENSE
├── requirements.txt
├── environment.yml
├── data/
│   ├── README.md
│   ├── metro/
│   ├── NHIS/
│   │   └── 2016~2017/
│   ├── mobility_factor/
│   │   └── 2016~2017/
│   ├── Rt/
│   │   ├── 2016~2017/
│   │   ├── 2017~2018/
│   │   ├── 2018~2019/
│   │   └── 2022~2023/
│   └── population_denominators_2016_2017.csv
├── notebooks/
│   ├── README.md
│   ├── 01_mobility_factor_2016_2017.ipynb
│   ├── 02_sir_negative_binomial_particle_smoothing.ipynb
│   ├── 03_rt_estimation_2016_2017.ipynb
│   ├── 04_rt_heatmap_scenarios.ipynb
│   └── 05_scenario_summary_tables.ipynb
├── figures/
│   ├── README.md
│   ├── mobility_factor/
│   ├── 2016~2017/
│   ├── HeatMap/
│   └── validation/
├── results/
│   ├── README.md
│   ├── tables/
│   │   └── scenario_summary_reference.csv
│   └── validation/
├── scripts/
│   ├── patch_notebook_paths.py
│   └── check_repository.py
└── docs/
    ├── code_inventory.md
    ├── data_sources.md
    ├── model_and_notation.md
    ├── output_manifest.md
    ├── region_definitions.md
    └── reproducibility_checklist.md

Installation

Clone the repository:

git clone https://github.com/jaewansh/KMOU-A-mobility-informed-SIR-model.git
cd KMOU-A-mobility-informed-SIR-model

Create a virtual environment:

python -m venv .venv
source .venv/bin/activate

On Windows:

.venv\Scripts\activate

Install dependencies:

python -m pip install --upgrade pip
pip install -r requirements.txt

Alternatively, create the conda environment:

conda env create -f environment.yml
conda activate mobility-informed-sir

Required input files

Subway mobility input

The mobility-factor notebook expects city-level daily subway files. Place them in:

data/metro/

Recommended file names:

seoul&g_metro.csv
busan_metro.csv
daejeon_metro.csv
daegu_metro.csv
gwangju_metro.csv

Required columns:

date
city
people_in

The analysis uses people_in, corresponding to subway boarding counts.

NHIS influenza input

The (R(t)) estimation notebook expects city-level daily influenza case files in:

data/NHIS/2016~2017/

Expected 2016-2017 file names:

Seoul&Gyeonggi_cases(2016~2017).csv
Busan_cases(2016~2017).csv
Daejeon_cases(2016~2017).csv
Daegu_cases(2016~2017).csv
Gwangju_cases(2016~2017).csv

The notebooks use diagnosis-code-based daily influenza incidence prepared from J09, J10, and J11 records.

Notebook path update

The uploaded notebooks originally contained local absolute paths from the author’s computer. After renaming the notebooks, run:

python scripts/patch_notebook_paths.py

Then verify that no local absolute paths remain:

python scripts/check_repository.py

Notebook execution order

Run the notebooks in this order.

1. Construct mobility factors

notebooks/01_mobility_factor_2016_2017.ipynb

Main outputs:

data/mobility_factor/2016~2017/Seoul_gamma.csv
data/mobility_factor/2016~2017/Busan_gamma.csv
data/mobility_factor/2016~2017/Daegu_gamma.csv
data/mobility_factor/2016~2017/Daejeon_gamma.csv
data/mobility_factor/2016~2017/Gwangju_gamma.csv

2. Validate particle smoothing under negative binomial observation

notebooks/02_sir_negative_binomial_particle_smoothing.ipynb

Main outputs:

results/validation/daily_rmse_results.csv
figures/validation/SIR Mathematical Model Negative binomial distribution RMSE.eps

3. Estimate (R(t)) for the 2016-2017 season

notebooks/03_rt_estimation_2016_2017.ipynb

Main outputs:

data/Rt/2016~2017/Seoul_Rt(2016~2017).csv
data/Rt/2016~2017/Busan_Rt(2016~2017).csv
data/Rt/2016~2017/Daegu_Rt(2016~2017).csv
data/Rt/2016~2017/Daejeon_Rt(2016~2017).csv
data/Rt/2016~2017/Gwangju_Rt(2016~2017).csv

4. Generate heatmaps

notebooks/04_rt_heatmap_scenarios.ipynb

This notebook reads (R(t)) files from:

data/Rt/2016~2017/
data/Rt/2017~2018/
data/Rt/2018~2019/
data/Rt/2022~2023/

and generates heatmaps for Seoul, Busan, Daegu, Daejeon, and Gwangju.

5. Generate scenario summary tables

notebooks/05_scenario_summary_tables.ipynb

This notebook computes mean (R(t)), reduction in mean (R(t)), days with (R(t)>1), reduction in days with (R(t)>1), and excess transmission area.

Output column conventions

For a city named City, the (R(t)) CSV files use the following columns:

Column	Meaning
`date`	Daily date
`City_Rt`	Baseline (R(t)) using observed mobility
`City_Rt_1`	(R(t)) under 10% mobility reduction
`City_Rt_2`	(R(t)) under 20% mobility reduction
`City_Rt_3`	(R(t)) under 30% mobility reduction
`City_Rt_4`	(R(t)) under 40% mobility reduction
`City_Rt_5`	(R(t)) under 50% mobility reduction

Reference scenario summary

A reference summary table transcribed from the manuscript and supplementary material is provided at:

results/tables/scenario_summary_reference.csv

This file includes the Seoul Table 2 values and Supplementary Tables S1-S4 for Busan, Daegu, Daejeon, and Gwangju.

License

The manuscript is distributed under the Creative Commons Attribution License as stated in the article.

The code repository is distributed under the license specified in the LICENSE file.

Contact

JAEWAN SHIN
Major of Data Science, Korea Maritime & Ocean University
Email: sjw6258@g.kmou.ac.kr

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

KMOU-A-mobility-informed-SIR-model

Overview

Important source-code notation

Study scope

Model summary

Mobility-factor construction

Mobility-reduction scenarios

Repository structure

Installation

Required input files

Subway mobility input

NHIS influenza input

Notebook path update

Notebook execution order

1. Construct mobility factors

2. Validate particle smoothing under negative binomial observation

3. Estimate (R(t)) for the 2016-2017 season

4. Generate heatmaps

5. Generate scenario summary tables

Output column conventions

Reference scenario summary

License

Contact