Machine learning for bioinformatics and neuroimaging

Abstract

Machine Learning (ML) is a well‐known paradigm that refers to the ability of systems to learn a specific task from the data and aims to develop computer algorithms that improve with experience. It involves computational methodologies to address complex real‐world problems and promises to enable computers to assist humans in the analysis of large, complex data sets. ML approaches have been widely applied to biomedical fields and a great body of research is devoted to this topic. The purpose of this article is to present the state‐of‐the art in ML applications to bioinformatics and neuroimaging and motivate research in new trend‐setting directions. We show how ML techniques such as clustering, classification, embedding techniques and network‐based approaches can be successfully employed to tackle various problems such as gene expression clustering, patient classification, brain networks analysis, and identification of biomarkers. We also present a short description of deep learning and multiview learning methodologies applied in these contexts. We discuss some representative methods to provide inspiring examples to illustrate how ML can be used to address these problems and how biomedical data can be characterized through ML. Challenges to be addressed and directions for future research are presented and an extensive bibliography is included. This article is categorized under: Application Areas > Health Care Technologies > Computational Intelligence Fundamental Concepts of Data and Knowledge > Motivation and Emergence of Data Mining Fundamental Concepts of Data and Knowledge > Key Design Issues in Data Mining

Transcriptomics in Toxicogenomics, Part III: Data Modelling for Risk Assessment

“…Dimensionality reduction techniques and feature selection methods can mitigate these issues and can be used in combination with ML approaches to build predictive modeling [24]. Some examples of dimensionality reduction techniques are principal component analysis (PCA) [88], multidimensional scaling (MDS) [89], t-distributed stochastic neighbour embedding (t-SNE) [90] and Uniform Manifold Approximation and Projection (UMAP) [91].…”

“…If the goal of the analysis is to reduce the dimensionality by preserving the original features, feature selection approaches can be a better alternative. Indeed, it allows to reveal significant underlying information and to identify a set of biomarkers for a particular phenotype [24]. Examples of these are filter approaches such as information gain, correlation feature selection (CFS) [93], Borda [94], random forests [95,96], FPRF [26], and Varsel [97].…”

“…This subset of features can then be used for the prediction of the class or the effect of a new sample. A wide range of algorithms has been proposed to build robust and accurate predictive models, including linear and logistic models, support vector machines (SVM), random forests (RF), classification and regression trees (CART), partial least squares discriminant analysis (PLSDA), linear discriminant analysis (LDA), artificial neural networks (ANNs), matrix factorization (MF) and k-nearest neighbours (K-NN) [23][24][25][26]. Classic techniques such as linear and logistic models have been the first to be applied in such modelling tasks and can still be considered the methods of choice, especially when analyzing small datasets.…”

Why High-Performance Modelling and Simulation for Big Data Applications Matters

“…Moreover, a specific survey on HTO is introduced by Suravajhala et al [122]. Nevertheless, a general survey oriented to high throughput biomedical data analysis with ML and DL is widely described in the work of Serra et al [123].…”

AB_SA: Tracing the source of bacterial strains based on accessory genes. Application toSalmonellaTyphimurium environmental strains

“…It is an extension of binary logistic regression allowing for more than two outcomes events. MLR has proven to provide a pertinent framework to carry out association analyses across multiple phenotypic traits (19; 20) and in foodborne outbreak investigations as a rule-out tool (21). Complex phenotypes, such as host adaptation in specific niches, have been linked to the presence of genes and genetic elements in some strains but not in others (referred as the “accessory genome”), mainly driven by horizontal DNA transfer (22; 7; 23).…”