Nucleic Acid Quantification by Multi-Frequency Impedance Cytometry and Machine Learning.

Area of Science:

• Molecular Diagnostics
• Biotechnology
• Machine Learning in Science

Background:

• Accurate nucleic acid quantification is crucial for molecular diagnostics and downstream analyses.
• Existing methods for DNA quantification can be limited, especially with small DNA input volumes.
• Novel, high-throughput, and accurate quantification methods are needed.

Purpose of the Study:

• To demonstrate the efficacy of a neural network for predicting DNA concentrations.
• To develop a machine learning model for DNA quantification using impedance data.
• To assess the performance of deep learning architectures for this application.

Main Methods:

• A custom microfluidic chip was utilized to detect DNA molecules bound to paramagnetic beads.
• Impedance peak response (IPR) was measured at multiple frequencies.
• Electrical measurements (frequency, imaginary and real parts of peak intensity) served as input for deep learning models.

Main Results:

• The study examined 10 different deep learning architectures.
• A regression model achieved an R-squared value of 97% and a slope of 0.68.
• The neural network successfully predicted DNA concentration from raw impedance data.

Conclusions:

• Machine learning models, particularly neural networks, provide a suitable, fast, and accurate method for nucleic acid concentration measurement.
• The proposed neural network effectively utilizes impedance data for DNA quantification.
• This approach has significant potential for applications in molecular diagnostics and high-throughput analysis.