Benchmarking feature selection and feature extraction methods to improve the performances of machine-learning algorithms for patient classification using metabolomics biomedical data

Objective

Classification tasks are an open challenge in the field of biomedicine. While several machine-learning techniques exist to accomplish this objective, several peculiarities associated with biomedical data, especially when it comes to omics measurements, prevent their use or good performance achievements. Omics approaches aim to understand a complex biological system through systematic analysis of its content at the molecular level. On the other hand, omics data are heterogeneous, sparse and affected by the classical "curse of dimensionality" problem, i.e. having much fewer observation, samples (n) than omics features (p). Furthermore, a major problem with multi-omics data is the imbalance either at the class or feature level. The objective of this work is to study whether feature extraction and/or feature selection techniques can improve the performances of classification machine-learning algorithms on omics measurements.

Methods

Among all omics, metabolomics has emerged as a powerful tool in cancer research, facilitating a deeper understanding of the complex metabolic landscape associated with tumorigenesis and tumor progression. Thus, we selected three publicly available metabolomics datasets, and we applied several feature extraction techniques both linear and non-linear, coupled or not with feature selection methods, and evaluated the performances regarding patient classification in the different configurations for the three datasets.

Results

We provide general workflow and guidelines on when to use those techniques depending on the characteristics of the data available. To further test the extension of our approach to other omics data, we have included a transcriptomics and a proteomics data. Overall, for all datasets, we showed that applying supervised feature selection improves the performances of feature extraction methods for classification purposes. Scripts used to perform all analyses are available at: https://github.com/Plant-Net/Metabolomic_project/.

ABEILLE: a novel method for ABerrant Expression Identification empLoying machine Learning from RNA-sequencing data

MOTIVATION

Current advances in omics technologies are paving the diagnosis of rare diseases proposing as a complementary assay to identify the responsible gene. The use of transcriptomic data to identify aberrant gene expression (AGE) have demonstrated to yield potential pathogenic events. However popular approaches for AGE identification are limited by the use of statistical tests that imply the choice of arbitrary cut-off for significance assessment and the availability of several replicates not always possible in clinical contexts.

RESULTS

Hence we developed ABEILLE (ABerrant Expression Identification empLoying machine LEarning from sequencing data) a variational autoencoder (VAE) based method for the identification of AGEs from the analysis of RNA-seq data without the need of replicates or a control group. ABEILLE combines the use of a VAE, able to model any data without specific assumptions on their distribution, and a decision tree to classify genes as AGE or non-AGE. An anomaly score is associated to each gene in order to stratify AGE by severity of aberration. We tested ABEILLE on semi-synthetic and an experimental dataset demonstrating the importance of the flexibility of the VAE configuration to identify potential pathogenic candidates.

AVAILABILITY

ABEILLE source code is freely available at : https://github.com/UCA-MSI/ABEILLE.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Multi-Omics Approaches to Improve Mitochondrial Disease Diagnosis: Challenges, Advances, and Perspectives

Mitochondrial diseases (MD) are rare disorders caused by deficiency of the mitochondrial respiratory chain, which provides energy in each cell. They are characterized by a high clinical and genetic heterogeneity and in most patients, the responsible gene is unknown. Diagnosis is based on the identification of the causative gene that allows genetic counseling, prenatal diagnosis, understanding of pathological mechanisms, and personalized therapeutic approaches. Despite the emergence of Next Generation Sequencing (NGS), to date, more than one out of two patients has no diagnosis in the absence of identification of the responsible gene. Technologies currently used for detecting causal variants (genetic alterations) is far from complete, leading many variants of unknown significance (VUS) and mainly based on the use of whole exome sequencing thus neglecting the identification of non-coding variants. The complexity of human genome and its regulation at multiple levels has led biologists to develop several assays to interrogate the different aspects of biological processes. While one-dimension single omics investigation offers a peek of this complex system, the combination of different omics data allows the discovery of coherent signatures. The community of computational biologists and bioinformaticians, in order to integrate data from different omics, has developed several approaches and tools. However, it is difficult to understand which suits the best to predict diverse phenotypic outcome. First attempts to use multi-omics approaches showed an improvement of the diagnostic power. However, we are far from a complete understanding of MD and their diagnosis. After reviewing multi-omics algorithms developed in the latest years, we are proposing here a novel data-driven classification and we will discuss how multi-omics will change and improve the diagnosis of MD. Due to the growing use of multi-omics approaches in MD, we foresee that this work will contribute to set up good practices to perform multi-omics data integration to improve the prediction of phenotypic outcomes and the diagnostic power of MD.