Clustering Alzheimer’s Disease Subtypes via Similarity Learning and Graph Diffusion

This repository holds the source code for the following manuscript accepted by ICIBM 2023:

@inproceedings{ADSIMLR2023,
  title={Clustering Alzheimer's Disease Subtypes via Similarity Learning and Graph Diffusion},
  author={Tianyi Wei and Shu Yang and Davoud Ataee Tarzanagh and Jingxuan Bao and Jia Xu and Patryk Orzechowski and Joost B. Wagenaar and Qi Long and Li Shen},
  booktitle={International Conference on Intelligent Biology and Medicine (ICIBM)},
  year={2023},
}

🦸‍ Abstract

Alzheimer's disease (AD) is a complex neurodegenerative disorder that affects millions of people worldwide. Due to the heterogeneous nature of AD, its diagnosis and treatment pose critical challenges. Consequently, there is a growing research interest in identifying homogeneous AD subtypes that can assist in addressing these challenges in recent years. In this study, we aim to identify subtypes of AD that represent distinctive clinical features and underlying pathology by utilizing unsupervised clustering with graph diffusion and similarity learning. We adopted SIMLR, a multi-kernel similarity learning framework, and graph diffusion to perform clustering on a group of 829 patients with AD and mild cognitive impairment (MCI, a prodromal stage of AD) based on their cortical thickness measurements extracted from magnetic resonance imaging (MRI) scans. Although the clustering approach we utilized has not been explored for the task of AD subtyping before, it demonstrated significantly better performance than several commonly used clustering methods. Specifically, we showed the power of graph diffusion in reducing the effects of noise in the subtype detection. Our results revealed five subtypes that differed remarkably in their biomarkers, cognitive status, and some other clinical features. To evaluate the resultant subtypes further, a genetic association study was carried out and successfully identified potential genetic underpinnings of different AD subtypes.

📝 Requirements

• Python 3.6 or later
• MATLAB 2021A or later

Install the required packages using pip:

pip install -r requirements.txt

🗄️ Data

The data used in the paper is a subset of the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. We are unable to provide the data used for this paper due to the presence of a nondisclosure agreement (NDA) that governs its use. The data contains sensitive or proprietary information that is subject to legal restrictions.

We understand the importance of transparency and reproducibility in research and development. While we are unable to share the actual dataset, we have provided the scripts and instructions needed to generate figures using a single-cell RNA-seq dataset (mECS) ¹ made to the public by the authors of SIMLR ², so you can understand the methodology and visualization techniques applied. This dataset contains 182 cells from 3 distinct cell populations. The data is stored in the "data" directory as "Test_1_mECS.mat".

🔨 Usage

Clone this repository

Clone this repository to your local machine:

git clone 
cd AD-SIMLR

The authors of SIMLR provided SIMLR code for R and MATLAB ³. We conducted our analysis using the MATLAB version of SIMLR.

Generate Clustering Results

The "SC_SIMLR_implementation.m" script contains code to generate Silhouette scores, similarity matrices, cluster assignments, and 2D visualization embeddings.

1. Open MATLAB on your machine.

2. Navigate to the directory "AD-SIMLR" containing the SC_SIMLR_implementation.m script.

3. Make sure you have the data in the "data" directory and specify the correct path in the script

   data_path = "data/Test_1_mECS.mat";
   

4. Run the script in MATLAB. The output will be saved in files under the results/ directory.

We modified the implementation in "SIMLR.m" to generate results before and after applying graph diffusion.

Generate Visualizations

The "generate_figures.py" script generates figures of similarity matrices and 2-D cluster visualization using data obtained from the MATLAB script. We used this script to generate Figure 3 and Figure 4 (kindly note that the cluster figures might not look exactly the same due to the stochastic nature of TSNE) in the paper.

python3 generate_figures.py --dir ./results

Results

We applied K-Means, spectral clustering, SIMLR, and the graph diffusion variants of the latter two methods to the mECS dataset. The Silhouette scores are stored in "./results/silhouette_scores.csv". We observed that the highest Silhouette score was achieved by SIMLR with graph diffusion.

	kmeans	SC	SC w/ diffusion	SIMLR	SIMLR w/ diffusion
Value	0.0566	0.7413	0.8723	0.5740	0.9835

Abbreviation: SC - spectral clustering

Below are the heatmaps of similarity matrices for spectral clustering, SIMLR, and their graph diffusion variants. All four similarity matrices have a block-diagonal structure. However, the similarity matrices of spectral clustering and SIMLR without graph diffusion suffered from a large amount of noise in the off diagonal entries (bottom row). After adding the graph diffusion, the noise in the similarity matrices were largely reduced (bottom row). The noise reduction effect of graph diffusion is especially remarkable under the similarity learning framework.

Spectral Clustering	SIMLR
Spectral Clustering + Graph Diffusion	SIMLR + Graph Diffusion

Below are the 2-D cluster visualizations for each of the five methods. The 2-D embeddings were obtained using t-SNE. Due to the stochastic nature of t-SNE, the 2-D visualizations may be different. We observed that SIMLR with graph diffusion resulted in the most well-separated visualization for the clusters.

K-Means	Spectral Clustering
SIMLR	Spectral Clustering + Graph Diffusion
SIMLR + Graph Diffusion

🤝 Acknowledgements

This work was supported in part by National Institutes of Health grants U01 AG068057, U01 AG066833, R01 AG071470, RF1 AG063481 and RF1 AG068191; and National Science Foundation grant IIS 1837964.

The data used in this repo was sourced from Buettner et al. 2015¹, and we thank them for providing valuable datasets. The SIMLR² framework was used extensively in the development of this code.

Authors and Affiliations

Department of Biostatistics, Epidemiology and Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA

Tianyi Wei#; Shu Yang#; Davoud Ataee Tarzanagh#; Jingxuan Bao; Jia Xu; Patryk Orzechowski; Joost B. Wagenaar; Qi Long; Li Shen

Department of Automatics and Robotics, AGH University of Science and Technology, al. Mickiewicza 30, 30-059 Krakow, Poland Patryk Orzechowski

📚 Citation

[1] Buettner, F. et al. Nat. Biotechnol. 33, 155–160 (2015).

[2] Wang B, Zhu J, Pierson E, Ramazzotti D, Batzoglou S. Visualization and analysis of single-cell RNA-seq data by kernel-based similarity learning. Nature methods. 2017;14(4):414-6.

[3] https://batzogloulabsu.github.io/SIMLR/