Machine-Learning-Based Single-Molecule Quantification of Circulating MicroRNA Mixtures

Jonathan Jeffet, Sayan Mondal, Amit Federbush, Nadav Tenenboim, Miriam Neaman, Jasline Deek, Yuval Ebenstein, Yohai Bar‐SinaiACS 2023 · DOI

TL;DR

This paper introduces MiRACLE, a novel machine learning-based method for simultaneously quantifying multiple circulating microRNAs (miRs) at the single-molecule level directly from blood plasma. This approach bypasses traditional sequencing and amplification steps, offering a faster, more cost-effective, and sensitive diagnostic tool for early disease detection.

What Problem Does This Paper Address?

Circulating microRNAs (miRs) are promising biomarkers for early disease detection, particularly in liquid biopsies. However, current methods for quantifying these small RNA molecules in complex biological samples like blood plasma face several challenges:

  • Limited Multiplexing: Many techniques struggle to simultaneously detect and quantify more than a few miR targets.
  • Amplification Bias: Standard methods like qRT-PCR rely on amplification, which can introduce biases for short RNA molecules like miRs.
  • Cost and Time: Techniques like RNA sequencing are expensive and time-consuming, especially for analyzing small panels of specific miRs.
  • Sensitivity and Specificity: While some methods offer multiplexing, they may lack the sensitivity or specificity required for accurate quantification.

The authors aim to address these limitations by developing a method that can directly detect and quantify multiple miRs simultaneously from a small blood sample, offering a more efficient and accurate diagnostic approach.

Methods

The MiRACLE (miR Analysis by spectral Classification LEarning) pipeline integrates several key components:

  • Targeted Capture and Labeling: Specific miRs are captured using custom DNA capture-reporter probes. Each probe is labeled with a unique pair of fluorophores, creating a "spectral barcode" for each miR. This allows for target recognition with single-nucleotide sensitivity.
  • Selective Immobilization: The hybridized miR-DNA probe complexes are selectively captured onto a glass surface using an anti-DNA-RNA hybrid antibody. This step, followed by extensive washing, removes excess probes and background noise, ensuring only target miRs are analyzed.
  • Compact Spectral Imaging (CoCoS): A novel imaging module, CoCoS, is used to capture fluorescent signals. This system uses prisms to spectrally disperse the light from the dual-fluorophore labels, creating a unique, spatially separated dual-spot pattern (Point Spread Function or PSF) for each miR. The distance and intensity distribution of these spots encode the miR's identity.
  • Machine Learning-Based Classification: A machine learning model, combining Principal Component Analysis (PCA) and Support Vector Machines (SVM), is trained to classify these spectral PSFs. The model learns to distinguish between different miR types and background noise based on their unique spectral signatures. This automated classification pipeline allows for the analysis of millions of individual miR signals.

Datasets

The paper describes the use of synthetic miR mixtures for training and benchmarking the MiRACLE pipeline. Specifically, they used mixtures of three target miRs (miR-15b-5p, miR-155-5p, and miR-126-3p) at various ratios (1:1:1 and 2:5:3). The authors also mention the development of a visual labeling tool (V-TIMDER) for creating training datasets.

Key Insights and Ideas

  • Spectral Barcoding for Multiplexing: The core innovation is using dual-fluorophore labeling to create unique "spectral barcodes" for each miR, enabling multiplexed detection without relying on sequencing.
  • Single-Molecule Sensitivity: By directly imaging and analyzing individual miR-probe complexes, the method achieves single-molecule sensitivity, crucial for detecting low-abundance biomarkers.
  • Machine Learning for Complex Signal Interpretation: The application of PCA-SVM to classify complex spectral PSFs is key to automating the identification and quantification of multiple miRs in a single image.
  • End-to-End Pipeline Efficiency: The integrated approach, from sample preparation to automated analysis, significantly reduces time and cost compared to traditional methods.
  • Potential for Broad Application: The method's ability to be extended to up to 11 miR targets suggests its potential for developing comprehensive diagnostic panels.

Key Results

  • The MiRACLE method successfully quantified and distinguished between synthetic mixtures of three target miRs (miR-15b-5p, miR-155-5p, and miR-126-3p) with high accuracy.
  • The machine learning classifier demonstrated strong performance, with minimal confusion between miR types and primarily misclassifying noise. For a 1:1:1 mixture, the classifier achieved high precision and recall for each miR type.
  • The system was able to accurately determine the ratio of miR targets in a 2:5:3 mixture, achieving a result very close to the expected ratio (2.10:4.95:2.95).
  • The authors estimate that the method is cost-effective, costing less than $4 per sample.
  • Simulations suggest the pipeline can be expanded to multiplex up to 11 miR targets.

Limitations and Future Work

  • Current Multiplexing Limit: While simulations suggest up to 11 miRs can be multiplexed, the current experimental validation focused on three miRs.
  • Axial Chromatic Aberrations: The paper notes slight differences between simulated and experimental PSFs due to axial chromatic aberrations, which were not fully accounted for in the simulations.
  • Potential for Other Error Sources: The authors acknowledge that other error sources not modeled in their analysis might exist.
  • Future Work: The authors suggest that the method could be easily expanded to multiplex more targets and that further development could involve introducing 3- and 4-fluorophore probes for even greater multiplexing capabilities. They also highlight the potential for this technology in clinical applications for profiling small panels of native miRs and other RNA targets.

Summary updated

Abstract

MicroRNAs (miRs) are small noncoding RNAs that regulate gene expression and are emerging as powerful indicators of diseases. MiRs are secreted in blood plasma and thus may report on systemic aberrations at an early stage via liquid biopsy analysis. We present a method for multiplexed single-molecule detection and quantification of a selected panel of miRs. The proposed assay does not depend on sequencing, requires less than 1 mL of blood, and provides fast results by direct analysis of native, unamplified miRs. This is enabled by a novel combination of compact spectral imaging and a machine learning-based detection scheme that allows simultaneous multiplexed classification of multiple miR targets per sample. The proposed end-to-end pipeline is extremely time efficient and cost-effective. We benchmark our method with synthetic mixtures of three target miRs, showcasing the ability to quantify and distinguish subtle ratio changes between miR targets.
BibTeX
@inproceedings{jeffet2023machine,
  author = {Jonathan Jeffet and Sayan Mondal and Amit Federbush and Nadav Tenenboim and Miriam Neaman and Jasline Deek and Yuval Ebenstein and Yohai Bar‐Sinai},
  title = {{Machine-Learning-Based Single-Molecule Quantification of Circulating MicroRNA Mixtures}},
  booktitle = {ACS},
  year = {2023},
  doi = {10.1021/acssensors.3c01234},
}
Figure