Accurate Blood-Based Diagnostic Biosignatures for Alzheimer’s Disease via Automated Machine Learning

Makrina Karaglani, Krystallia Gourlia, Ioannis Tsamardinos, Εkaterini ChatzakiJournal of Clinical Medicine 2020 · DOI

TL;DR

This paper presents a novel approach to identifying blood-based biosignatures for Alzheimer's Disease (AD) using automated machine learning (AutoML). By leveraging publicly available multi-omics data, the researchers developed accurate predictive models that could pave the way for minimally invasive diagnostic tests for AD.

What Problem Does This Paper Address?

Diagnosing Alzheimer's Disease (AD) early and accurately remains a significant challenge. Current diagnostic methods often rely on clinical assessments, neuroimaging, and neuropsychological evaluations, with definitive diagnosis typically only possible post-mortem. The lack of reliable, minimally invasive blood-based biomarkers hinders early intervention and effective disease management. This paper aims to address this gap by exploring the potential of multi-omics data (proteomics, transcriptomics, and metabolomics) from blood samples to identify robust diagnostic biosignatures for AD.

Methods

The core of this study lies in the application of an AutoML platform called "Just Add Data Bio" (JADBIO). JADBIO is designed to automate the process of building predictive models from high-dimensional, low-sample omics data. Its key features include:

  • Automated Algorithm Selection: JADBIO automatically selects and tunes appropriate classification algorithms (e.g., Support Vector Machines, Random Forests, Ridge Logistic Regression) without requiring manual intervention.
  • Feature Selection: It employs methods like Statistically Equivalent Biosignatures (SES) to identify the minimal set of features (biomarkers) that yield the highest predictive power.
  • Robust Performance Estimation: The platform uses bootstrapping techniques (specifically, Bootstrap Bias Corrected cross-validation or BBC-CV) to provide conservative and reliable estimates of out-of-sample performance, mitigating the risk of over-optimistic results.
  • Multi-Omics Integration: JADBIO can handle various omics data types, allowing for the exploration of different molecular profiles.

The researchers used JADBIO to analyze multiple publicly available datasets, aiming to construct predictive models for AD diagnosis.

Datasets

The study utilized several publicly available multi-omics datasets from blood samples of AD patients and cognitively healthy individuals. These datasets were sourced from repositories like BioDataome, Metabolomics Workbench, and Gene Expression Omnibus (GEO). Specifically, the following types of datasets were analyzed:

  • Proteomic datasets: Two datasets were pooled for analysis.
  • miRNA transcriptomic datasets: Two datasets were used.
  • mRNA transcriptomic datasets: Four datasets were used.
  • Metabolomic datasets: Two datasets were used.

The paper notes that some datasets were excluded due to very low sample numbers or lack of processed data. The study also mentions internal validation by splitting larger datasets into 70/30 train/test ratios, where applicable.

Key Insights and Ideas

  • AutoML as a powerful tool for biomarker discovery: The study strongly highlights the value of AutoML platforms like JADBIO in democratizing complex data analysis. It allows researchers without extensive coding or statistical expertise to generate high-quality predictive models, accelerating the biomarker discovery process.
  • The potential of blood-based omics for AD diagnosis: The successful identification of accurate biosignatures from blood samples underscores the feasibility of developing non-invasive diagnostic tests for AD. This is a significant step towards earlier and more accessible diagnosis.
  • The importance of re-analyzing existing data: The paper emphasizes that "old" datasets can still yield novel insights when analyzed with new, advanced computational methods. This challenges the notion that all discoveries have been made from existing data.
  • Reduced biosignature size is crucial for clinical translation: The development of biosignatures with a small number of predictors (e.g., three miRNAs or seven proteins) is highlighted as a key advantage for cost-effective and easily implementable clinical assays.

Key Results

The study successfully developed three high-performing diagnostic biosignatures for AD:

  • miRNA-based biosignature: Using a 506-feature miRNA transcriptomic dataset, a Support Vector Machine model identified three miRNA predictors with a high Area Under the Curve (AUC) of 0.975 (0.906, 1.000).
  • mRNA-based biosignature: From a 38,327-feature mRNA transcriptomic dataset, six statistically equivalent Classification Random Forests models were developed, each using 25 mRNA predictors, achieving an AUC of 0.846 (0.778, 0.905).
  • Protein-based biosignature: Analyzing a 9483-feature proteomic dataset, a Ridge Logistic Regression model identified seven protein predictors with an AUC of 0.921 (0.849, 0.972).

These performance metrics were validated through JADBIO's internal cross-validation, demonstrating stability. The paper also notes that the identified miRNAs and proteins showed varying degrees of relation to AD according to GeneCards searches, with some proteins being newly implicated.

Limitations and Future Work

  • Absence of external validation: A significant limitation acknowledged by the authors is the lack of external validation using independent datasets run on common platforms. While internal cross-validation was performed, this limits the generalizability of the findings.
  • Limited clinical information: The available datasets primarily contained age and sex information, with no details on disease stage, duration, progression, or specific dementia subtypes. This restricts the diagnostic scope of the developed biosignatures to differentiating AD from healthy controls.
  • Small sample sizes for proteomics and metabolomics: The proteomic and metabolomic datasets had relatively low sample numbers, which the authors note was partially overcome by the AutoML approach.
  • Future Work: The authors strongly advocate for clinical validation of these biosignatures using respective laboratory assays. They also suggest that future research could explore multi-omic approaches and investigate the newly implicated proteins in more detail. The potential for these biosignatures to be translated into cost-effective, routine blood tests in standard diagnostic labs is a key future direction.

Summary updated

Abstract

Alzheimer's disease (AD) is the most common form of neurodegenerative dementia and its timely diagnosis remains a major challenge in biomarker discovery. In the present study, we analyzed publicly available high-throughput low-sample -omics datasets from studies in AD blood, by the AutoML technology Just Add Data Bio (JADBIO), to construct accurate predictive models for use as diagnostic biosignatures. Considering data from AD patients and age-sex matched cognitively healthy individuals, we produced three best performing diagnostic biosignatures specific for the presence of AD: A. A 506-feature transcriptomic dataset from 48 AD and 22 controls led to a miRNA-based biosignature via Support Vector Machines with three miRNA predictors (AUC 0.975 (0.906, 1.000)), B. A 38,327-feature transcriptomic dataset from 134 AD and 100 controls led to six mRNA-based statistically equivalent signatures via Classification Random Forests with 25 mRNA predictors (AUC 0.846 (0.778, 0.905)) and C. A 9483-feature proteomic dataset from 25 AD and 37 controls led to a protein-based biosignature via Ridge Logistic Regression with seven protein predictors (AUC 0.921 (0.849, 0.972)). These performance metrics were also validated through the JADBIO pipeline confirming stability. In conclusion, using the automated machine learning tool JADBIO, we produced accurate predictive biosignatures extrapolating available low sample -omics data. These results offer options for minimally invasive blood-based diagnostic tests for AD, awaiting clinical validation based on respective laboratory assays. They also highlight the value of AutoML in biomarker discovery.
BibTeX
@article{karaglani2020accurate,
  author = {Makrina Karaglani and Krystallia Gourlia and Ioannis Tsamardinos and Εkaterini Chatzaki},
  title = {{Accurate Blood-Based Diagnostic Biosignatures for Alzheimer’s Disease via Automated Machine Learning}},
  journal = {Journal of Clinical Medicine},
  year = {2020},
  doi = {10.3390/jcm9093016},
}
Figure