What Can Machine Learning Approaches in Genomics Tell Us about the Molecular Basis of Amyotrophic Lateral Sclerosis?

AI-enabled evaluation of genome-wide association relevance and polygenic risk score prediction in Alzheimer's disease

GWAS focuses on significance loosing false positives; machine learning probes sub-significant features relying on predictivity. Yet, these are far from orthogonal. We sought to explore how these inform each other in sub-genome-wide significant situations to define relevance for predictive features. We introduce the SVM-based RubricOE that selects heavily cross-validated feature sets, and LDpred2 PRS as a strong contrast to SVM, to explore significance and predictivity. Our Alzheimer's test case notoriously lacks strong genetic signals except for few very strong phenotype-SNP associations, which suits the problem we are exploring. We found that the most significant SNPs among ML and PRS-selected SNPs captured most of the predictivity, while weaker associations tend also to contribute weakly to predictivity. SNPs with weak associations tend not to contribute to predictivity, but deletion of these features does not injure it. Significance provides a ranking that helps identify weakly predictive features.

Genome-Wide Gene-Set Analysis Identifies Molecular Mechanisms Associated with ALS

Amyotrophic lateral sclerosis (ALS) is a fatal late-onset motor neuron disease characterized by the loss of the upper and lower motor neurons. Our understanding of the molecular basis of ALS pathology remains elusive, complicating the development of efficient treatment. Gene-set analyses of genome-wide data have offered insight into the biological processes and pathways of complex diseases and can suggest new hypotheses regarding causal mechanisms. Our aim in this study was to identify and explore biological pathways and other gene sets having genomic association to ALS. Two cohorts of genomic data from the dbGaP repository were combined: (a) the largest available ALS individual-level genotype dataset (N = 12,319), and (b) a similarly sized control cohort (N = 13,210). Following comprehensive quality control pipelines, imputation and meta-analysis, we assembled a large European descent ALS-control cohort of 9244 ALS cases and 12,795 healthy controls represented by genetic variants of 19,242 genes. Multi-marker analysis of genomic annotation (MAGMA) gene-set analysis was applied to an extensive collection of 31,454 gene sets from the molecular signatures database (MSigDB). Statistically significant associations were observed for gene sets related to immune response, apoptosis, lipid metabolism, neuron differentiation, muscle cell function, synaptic plasticity and development. We also report novel interactions between gene sets, suggestive of mechanistic overlaps. A manual meta-categorization and enrichment mapping approach is used to explore the overlap of gene membership between significant gene sets, revealing a number of shared mechanisms.

Gene targeting in amyotrophic lateral sclerosis using causality-based feature selection and machine learning

BACKGROUND

Amyotrophic lateral sclerosis (ALS) is a rare progressive neurodegenerative disease that affects upper and lower motor neurons. As the molecular basis of the disease is still elusive, the development of high-throughput sequencing technologies, combined with data mining techniques and machine learning methods, could provide remarkable results in identifying pathogenetic mechanisms. High dimensionality is a major problem when applying machine learning techniques in biomedical data analysis, since a huge number of features is available for a limited number of samples. The aim of this study was to develop a methodology for training interpretable machine learning models in the classification of ALS and ALS-subtypes samples, using gene expression datasets.

METHODS

We performed dimensionality reduction in gene expression data using a semi-automated preprocessing systematic gene selection procedure using Statistically Equivalent Signature (SES), a causality-based feature selection algorithm, followed by Boosted Regression Trees (XGBoost) and Random Forest to train the machine learning classifiers. The SHapley Additive exPlanations (SHAP values) were used for interpretation of the machine learning classifiers. The methodology was developed and tested using two distinct publicly available ALS RNA-seq datasets. We evaluated the performance of SES as a dimensionality reduction method against: (a) Least Absolute Shrinkage and Selection Operator (LASSO), and (b) Local Outlier Factor (LOF).

RESULTS

The proposed methodology achieved 85.18% accuracy for the classification of cerebellum or frontal cortex samples as C9orf72-related familial ALS, sporadic ALS or healthy samples. Importantly, the genes identified as the most determinative have also been reported as disease-associated in ALS literature. When tested in the evaluation dataset, the methodology achieved 88.89% accuracy for the classification of sporadic ALS motor neuron samples. When LASSO was used as feature selection method instead of SES, the accuracy of the machine learning classifiers ranged from 74.07 to 96.30%, depending on tissue assessed, while LOF underperformed significantly (77.78% accuracy for the classification of pooled cerebellum and frontal cortex samples).

CONCLUSIONS

Using SES, we addressed the challenge of high dimensionality in gene expression data analysis, and we trained accurate machine learning ALS classifiers, specific for the gene expression patterns of different disease subtypes and tissue samples, while identifying disease-associated genes.

Machine learning methods applied to genotyping data capture interactions between single nucleotide variants in late onset Alzheimer's disease

Introduction

Genome-wide association studies (GWAS) in late onset Alzheimer's disease (LOAD) provide lists of individual genetic determinants. However, GWAS do not capture the synergistic effects among multiple genetic variants and lack good specificity.

Methods

We applied tree-based machine learning algorithms (MLs) to discriminate LOAD (>700 individuals) and age-matched unaffected subjects in UK Biobank with single nucleotide variants (SNVs) from Alzheimer's disease (AD) studies, obtaining specific genomic profiles with the prioritized SNVs.

Results

MLs prioritized a set of SNVs located in genes PVRL2, TOMM40, APOE, and APOC1, also influencing gene expression and splicing. The genomic profiles in this region showed interaction patterns involving rs405509 and rs1160985, also present in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. rs405509 located in APOE promoter interacts with rs429358 among others, seemingly neutralizing their predisposing effect.

Discussion

Our approach efficiently discriminates LOAD from controls, capturing genomic profiles defined by interactions among SNVs in a hot-spot region.

Optimized Molecular Interaction Networks for the Study of Skeletal Muscle

BACKGROUND

Molecular interaction networks (MINs) aim to capture the complex relationships between interacting molecules within a biological system. MINs can be constructed from existing knowledge of molecular functional associations, such as protein-protein binding interactions (PPI) or gene co-expression, and these different sources may be combined into a single MIN. A given MIN may be more or less optimal in its representation of the important functional relationships of molecules in a tissue.

OBJECTIVE

The aim of this study was to establish whether a combined MIN derived from different types of functional association could better capture muscle-relevant biology compared to its constituent single-source MINs.

METHODS

MINs were constructed from functional association databases for both protein-binding and gene co-expression. The networks were then compared based on the capture of muscle-relevant genes and gene ontology (GO) terms, tested in two different ways using established biological network clustering algorithms. The top performing MINs were combined to test whether an optimal MIN for skeletal muscle could be constructed.

RESULTS

The STRING PPI network was the best performing single-source MIN among those tested. Combining STRING with interactions from either the MyoMiner or CoXPRESSdb gene co-expression sources resulted in a combined network with improved performance relative to its constituent networks.

CONCLUSION

MINs constructed from multiple types of functional association can better represent the functional relationships of molecules in a given tissue. Such networks may be used to improve the analysis and interpretation of functional genomics data in the study of skeletal muscle and neuromuscular diseases. Networks and clusters described by this study, including the combinations of STRING with MyoMiner or with CoXPRESSdb, are available for download from https://www.sys-myo.com/myominer/download.php.

snpQT: flexible, reproducible, and comprehensive quality control and imputation of genomic data

Quality control of genomic data is an essential but complicated multi-step procedure, often requiring separate installation and expert familiarity with a combination of different bioinformatics tools. Software incompatibilities, and inconsistencies across computing environments, are recurrent challenges, leading to poor reproducibility. Existing semi-automated or automated solutions lack comprehensive quality checks, flexible workflow architecture, and user control. To address these challenges, we have developed snpQT: a scalable, stand-alone software pipeline using nextflow and BioContainers, for comprehensive, reproducible and interactive quality control of human genomic data. snpQT offers some 36 discrete quality filters or correction steps in a complete standardised pipeline, producing graphical reports to demonstrate the state of data before and after each quality control procedure. This includes human genome build conversion, population stratification against data from the 1,000 Genomes Project, automated population outlier removal, and built-in imputation with its own pre- and post- quality controls. Common input formats are used, and a synthetic dataset and comprehensive online tutorial are provided for testing, educational purposes, and demonstration. The snpQT pipeline is designed to run with minimal user input and coding experience; quality control steps are implemented with numerous user-modifiable thresholds, and workflows can be flexibly combined in custom combinations. snpQT is open source and freely available at https://github.com/nebfield/snpQT. A comprehensive online tutorial and installation guide is provided through to GWAS (https://snpqt.readthedocs.io/en/latest/), introducing snpQT using a synthetic demonstration dataset and a real-world Amyotrophic Lateral Sclerosis SNP-array dataset.

snpQT: flexible, reproducible, and comprehensive quality control and imputation of genomic data

Quality control of genomic data is an essential but complicated multi-step procedure, often requiring separate installation and expert familiarity with a combination of different bioinformatics tools. Dependency hell and reproducibility are recurrent challenges. Existing semi-automated or automated solutions lack comprehensive quality checks, flexible workflow architecture, and user control. To address these challenges, we have developed snpQT: a scalable, stand-alone software pipeline using nextflow and BioContainers, for comprehensive, reproducible and interactive quality control of human genomic data. snpQT offers some 36 discrete quality filters or correction steps in a complete standardised pipeline, producing graphical reports to demonstrate the state of data before and after each quality control procedure. This includes human genome build conversion, population stratification against data from the 1,000 Genomes Project, automated population outlier removal, and built-in imputation with its own pre- and post- quality controls. Common input formats are used, and a synthetic dataset and comprehensive online tutorial are provided for testing, educational purposes, and demonstration. The snpQT pipeline is designed to run with minimal user input and coding experience; quality control steps are implemented with default thresholds which can be modified by the user, and workflows can be flexibly combined in custom combinations. snpQT is open source and freely available at https://github.com/nebfield/snpQT. A comprehensive online tutorial and installation guide is provided through to GWAS (https://snpqt.readthedocs.io/en/latest/), introducing snpQT using a synthetic demonstration dataset and a real-world Amyotrophic Lateral Sclerosis SNP-array dataset.

Amyotrophic Lateral Sclerosis and Frontotemporal Lobar Degenerations: Similarities in Genetic Background

Amyotrophic lateral sclerosis (ALS) is a devastating, uniformly lethal progressive degenerative disorder of motor neurons that overlaps with frontotemporal lobar degeneration (FTLD) clinically, morphologically, and genetically. Although many distinct mutations in various genes are known to cause amyotrophic lateral sclerosis, it remains poorly understood how they selectively impact motor neuron biology and whether they converge on common pathways to cause neuronal degeneration. Many of the gene mutations are in proteins that share similar functions. They can be grouped into those associated with cell axon dynamics and those associated with cellular phagocytic machinery, namely protein aggregation and metabolism, apoptosis, and intracellular nucleic acid transport. Analysis of pathways implicated by mutant ALS genes has provided new insights into the pathogenesis of both familial forms of ALS (fALS) and sporadic forms (sALS), although, regrettably, this has not yet yielded definitive treatments. Many genes play an important role, with TARDBP, SQSTM1, VCP, FUS, TBK1, CHCHD10, and most importantly, C9orf72 being critical genetic players in these neurological disorders. In this mini-review, we will focus on the molecular mechanisms of these two diseases.

Bayesian Network as a Decision Tool for Predicting ALS Disease

Clinical diagnosis of amyotrophic lateral sclerosis (ALS) is difficult in the early period. But blood tests are less time consuming and low cost methods compared to other methods for the diagnosis. The ALS researchers have been used machine learning methods to predict the genetic architecture of disease. In this study we take advantages of Bayesian networks and machine learning methods to predict the ALS patients with blood plasma protein level and independent personal features. According to the comparison results, Bayesian Networks produced best results with accuracy (0.887), area under the curve (AUC) (0.970) and other comparison metrics. We confirmed that sex and age are effective variables on the ALS. In addition, we found that the probability of onset involvement in the ALS patients is very high. Also, a person’s other chronic or neurological diseases are associated with the ALS disease. Finally, we confirmed that the Parkin level may also have an effect on the ALS disease. While this protein is at very low levels in Parkinson’s patients, it is higher in the ALS patients than all control groups.