DMT-EV: An Explainable Deep Network for Dimension Reduction

Abstract

Dimension reduction (DR) is commonly utilized to capture intrinsic structure and transform high-dimensional data into low-dimensional space while retaining meaningful properties of original data. It is used in a wide variety of applications such as image recognition, single-cell sequencing analysis, and biomarker discovery. However, contemporary parametric-free and parametric DR techniques suffer from several major shortcomings, such as the inability to preserve both global and local features and the pool generalization performance. On the other hand, regarding explainability, it is crucial to comprehend the embedding process, especially the contribution of each feature to the embedding process while understanding how each feature affects the embedding results that identify critical components and helps diagnose the embedding process. To address these problems, we have developed a deep neural network method called EVNet which provides not only excellent performance in structural maintainability but also explainability to the DR therein. EVNet starts from data augmentation and with a manifold-based loss function to improve embedding performance. The explanation is based on saliency maps and is aimed to examine the trained EVNet parameters and contributions of components during the embedding process. The proposed techniques are integrated with a visual interface to help the user to adjust EVNet to achieve better DR performance and explainability. The interactive visual interface makes it easier to illustrate the data features, compare different DR techniques, and investigate the explainability of DR. An in-depth experimental comparison is provided which shows that EVNet consistently outperforms the state-of-the-art methods in both performance measures and explainability.

BUPNN: Manifold Learning Regularizer‐Based Blood Usage Prediction Neural Network for Blood Centers

“…Recently, ML algorithms have been successfully used to classify thyroid cancer [7] and predict the progression of COVID-19 [8,9]. On the other hand, ML-based visualization and dimensionality reduction techniques have the potential to help professionals analyze biological or medical data, guiding them to better understand the data [10,11]. Furthermore, ML-based feature selection techniques [12,13] have strong interpretability and the potential to fnd highly relevant biomarkers for output in a wide range of medical data, leading to new biological or medical discoveries.…”

“…L D ⟵ D(S y , S z ) ; ⊳ # calculate loss function in equation ( 10). (10) θ ⟵ θ − ηzL D /zθ, ϕ ⟵ ϕ − ηzL D /zϕ; ⊳ # update parameters. (11) end while (12) end while (13)…”

Structure-preserving visualization for single-cell RNA-Seq profiles using deep manifold transformation with batch-correction

“…In this paper, we address the following application scenarios: Firstly, we develop a machine learning method with the ability to preserve geometric structure of the high dimensional scRNA-seq data in the dimensionality reduced space and visualization of scRNA-seq data that can be applied to both cell clustering and trajectory inference tasks. These two scenarios are closely related yet have different technical goals: (1) For cell clustering is to explore the relationship between different cell types at a given time [4][5][6][7][8][9][10][11][12][13][14][15][16][17] , which we call the static (at a time point) scenario. It is to learn a low-dimensional embedding in which cells belonging to the same type should be close to each other whereas those of different types be away from each other.…”

“…In recent years, deep neural networks (DNNs) 23 have been utilized as effective non-linear dimensionality reduction and visualization tools for processing large datasets, incorporating different factors, and improving the scalable ability of models. This field mainly involves two mainstream directions, including (1) Deep manifold learning methods, such as parametric UMAP 10 , Markov-Lipschitz deep learning (MLDL) 11 , deep manifold transformation (DMT) 12 , deep local-flatness manifold embedding (DLME) 24 , EVNet 13 , unified dimensional reduction neural-network (UDRN) 14 and IVIS 15 , and (2) Deep reconstruction learning methods, which covers various (variational) autoencoders 16,25,26 . Generally speaking, the latter seeks to reconstruct the input data distribution and often ignores the importance of intrinsic geometric structure in input data.…”