Careful feature selection is key in classification of Alzheimer's disease patients based on whole-genome sequencing data

Genomic Machine Learning Meta-regression: Insights on Associations of Study Features With Reported Model Performance

Many studies have been conducted with the goal of correctly predicting diagnostic status of a disorder using the combination of genomic data and machine learning. It is often hard to judge which components of a study led to better results and whether better reported results represent a true improvement or an uncorrected bias inflating performance. We extracted information about the methods used and other differentiating features in genomic machine learning models. We used these features in linear regressions predicting model performance. We tested for univariate and multivariate associations as well as interactions between features. Of the models reviewed, 46% used feature selection methods that can lead to data leakage. Across our models, the number of hyperparameter optimizations reported, data leakage due to feature selection, model type, and modeling an autoimmune disorder were significantly associated with an increase in reported model performance. We found a significant, negative interaction between data leakage and training size. Our results suggest that methods susceptible to data leakage are prevalent among genomic machine learning research, resulting in inflated reported performance. Best practice guidelines that promote the avoidance and recognition of data leakage may help the field avoid biased results.

Epistatic Features and Machine Learning Improve Alzheimer's Disease Risk Prediction Over Polygenic Risk Scores

Background

Polygenic risk scores (PRS) are linear combinations of genetic markers weighted by effect size that are commonly used to predict disease risk. For complex heritable diseases such as late-onset Alzheimer's disease (LOAD), PRS models fail to capture much of the heritability. Additionally, PRS models are highly dependent on the population structure of the data on which effect sizes are assessed and have poor generalizability to new data.

Objective

The goal of this study is to construct a paragenic risk score that, in addition to single genetic marker data used in PRS, incorporates epistatic interaction features and machine learning methods to predict risk for LOAD.

Methods

We construct a new state-of-the-art genetic model for risk of Alzheimer's disease. Our approach innovates over PRS models in two ways: First, by directly incorporating epistatic interactions between SNP loci using an evolutionary algorithm guided by shared pathway information; and second, by estimating risk via an ensemble of non-linear machine learning models rather than a single linear model. We compare the paragenic model to several PRS models from the literature trained on the same dataset.

Results

The paragenic model is significantly more accurate than the PRS models under 10-fold cross-validation, obtaining an AUC of 83% and near-clinically significant matched sensitivity/specificity of 75%. It remains significantly more accurate when evaluated on an independent holdout dataset and maintains accuracy within APOE genotype strata.

Conclusions

Paragenic models show potential for improving disease risk prediction for complex heritable diseases such as LOAD over PRS models.

The fusiform gyrus exhibits differential gene-gene co-expression in Alzheimer's disease

Alzheimer's Disease (AD) is an irreversible neurodegenerative disease clinically characterized by the presence of β-amyloid plaques and tau deposits in various regions of the brain. However, the underlying factors that contribute to the development of AD remain unclear. Recently, the fusiform gyrus has been identified as a critical brain region associated with mild cognitive impairment, which may increase the risk of AD development. In our study, we performed gene co-expression and differential co-expression network analyses, as well as gene-expression-based prediction, using RNA-seq transcriptome data from post-mortem fusiform gyrus tissue samples collected from both cognitively healthy individuals and those with AD. We accessed differential co-expression networks in large cohorts such as ROSMAP, MSBB, and Mayo, and conducted over-representation analyses of gene pathways and gene ontology. Our results comprise four exclusive gene hubs in co-expression modules of Alzheimer's Disease, including FNDC3A, MED23, NRIP1, and PKN2. Further, we identified three genes with differential co-expressed links, namely FAM153B, CYP2C8, and CKMT1B. The differential co-expressed network showed moderate predictive performance for AD, with an area under the curve ranging from 0.71 to 0.76 (+/- 0.07). The over-representation analysis identified enrichment for Toll-Like Receptors Cascades and signaling pathways, such as G protein events, PIP2 hydrolysis and EPH-Epherin mechanism, in the fusiform gyrus. In conclusion, our findings shed new light on the molecular pathophysiology of AD by identifying new genes and biological pathways involved, emphasizing the crucial role of gene regulatory networks in the fusiform gyrus.

Wide and deep learning based approaches for classification of Alzheimer's disease using genome-wide association studies

The increasing incidence of Alzheimer's disease (AD) has been leading towards a significant growth in socioeconomic challenges. A reliable prediction of AD might be useful to mitigate or at-least slow down its progression for which, identification of the factors affecting the AD and its accurate diagnoses, are vital. In this study, we use Genome-Wide Association Studies (GWAS) dataset which comprises significant genetic markers of complex diseases. The original dataset contains large number of attributes (620901) for which we propose a hybrid feature selection approach based on association test, principal component analysis, and the Boruta algorithm, to identify the most promising predictors of AD. The selected features are then forwarded to a wide and deep neural network models to classify the AD cases and healthy controls. The experimental outcomes indicate that our approach outperformed the existing methods when evaluated on standard dataset, producing an accuracy and f1-score of 99%. The outcomes from this study are impactful particularly, the identified features comprising AD-associated genes and a reliable classification model that might be useful for other chronic diseases.

Epistatic Features and Machine Learning Improve Alzheimer's Risk Prediction Over Polygenic Risk Scores

Background

Polygenic risk scores (PRS) are linear combinations of genetic markers weighted by effect size that are commonly used to predict disease risk. For complex heritable diseases such as late onset Alzheimer's disease (LOAD), PRS models fail to capture much of the heritability. Additionally, PRS models are highly dependent on the population structure of data on which effect sizes are assessed, and have poor generalizability to new data.

Objective

The goal of this study is to construct a paragenic risk score that, in addition to single genetic marker data used in PRS, incorporates epistatic interaction features and machine learning methods to predict lifetime risk for LOAD.

Methods

We construct a new state-of-the-art genetic model for lifetime risk of Alzheimer's disease. Our approach innovates over PRS models in two ways: First, by directly incorporating epistatic interactions between SNP loci using an evolutionary algorithm guided by shared pathway information; and second, by estimating risk via an ensemble of machine learning models (gradient boosting machines and deep learning) instead of simple logistic regression. We compare the paragenic model to a PRS model from the literature trained on the same dataset.

Results

The paragenic model is significantly more accurate than the PRS model under 10-fold cross-validation, obtaining an AUC of 83% and near-clinically significant matched sensitivity/specificity of 75%, and remains significantly more accurate when evaluated on an independent holdout dataset. Additionally, the paragenic model maintains accuracy within APOE genotypes.

Conclusion

Paragenic models show potential for improving lifetime disease risk prediction for complex heritable diseases such as LOAD over PRS models.

The Polygenic Risk Score Knowledge Base offers a centralized online repository for calculating and contextualizing polygenic risk scores

The process of identifying suitable genome-wide association (GWA) studies and formatting the data to calculate multiple polygenic risk scores on a single genome can be laborious. Here, we present a centralized polygenic risk score calculator currently containing over 250,000 genetic variant associations from the NHGRI-EBI GWAS Catalog for users to easily calculate sample-specific polygenic risk scores with comparable results to other available tools. Polygenic risk scores are calculated either online through the Polygenic Risk Score Knowledge Base (PRSKB; https://prs.byu.edu ) or via a command-line interface. We report study-specific polygenic risk scores across the UK Biobank, 1000 Genomes, and the Alzheimer's Disease Neuroimaging Initiative (ADNI), contextualize computed scores, and identify potentially confounding genetic risk factors in ADNI. We introduce a streamlined analysis tool and web interface to calculate and contextualize polygenic risk scores across various studies, which we anticipate will facilitate a wider adaptation of polygenic risk scores in future disease research.