Generalized EmbedSOM on quadtree-structured self-organizing maps

Miroslav Kratochvíl,Abhishek Koladiya,Jiří Vondrášek

doi:10.12688/f1000research.21642.2

Abstract

EmbedSOM is a simple and fast dimensionality reduction algorithm, originally developed for its applications in single-cell cytometry data analysis. We present an updated version of EmbedSOM, viewed as an algorithm for landmark-directed embedding enrichment, and demonstrate that it works well even with manifold-learning techniques other than the self-organizing maps. Using this generalization, we introduce an inwards-growing variant of self-organizing maps that is designed to mitigate some earlier identified deficiencies of EmbedSOM output. Finally, we measure the performance of the generalized EmbedSOM, compare several variants of the algorithm that utilize different landmark-generating functions, and showcase the functionality on single-cell cytometry datasets from recent studies.

Highlights

EmbedSOM is a dimensionality reduction (DR) algorithm for single-cell cytometry data, designed for high scalability, computational efficiency and performance[1]
R EV IS E D Amendments from Version 1. This version improves upon the main issues raised by the reviewers: We have added an useful comparison with other dimensionality reduction methods, and a slightly technical overview of the differences in a separate section
We focus on fixing inconsistencies and problems of the first version of EmbedSOM: First, we describe an updated version of EmbedSOM that improves the approximation to achieve mathematical smoothness of the projection

Summary

Nikolay Oskolkov Sweden

Any reports and responses or comments on the article can be found at the end of the article. This version improves upon the main issues raised by the reviewers: We have added an useful comparison with other dimensionality reduction methods (results on a toy dataset can be compared, performance of the EmbedSOM implementation is compared with UMAP, tSNE and TriMap in Figure 2), and a slightly technical overview of the differences in a separate section. We have fixed several wording problems and corrected minor mistakes and difficulties throughout the text, mainly in the description of Wong dataset (mainly providing a cleaner explanation of the phenomenon with γδTCR T cells, noticed by the reviewers). Any further responses from the reviewers can be found at the end of the article

Introduction

Methods

Van Der Maaten L

Borodin PA

12. Samet H