Machine Learning Prediction of DNA Charge Transport

Area of Science:

• Computational chemistry
• Materials science
• Molecular electronics

Background:

• Predicting charge transport in DNA is computationally intensive due to atomic motion.
• Screening DNA sequences for electrical conductivity is challenging with current methods.

Purpose of the Study:

• To develop a machine learning model for rapid prediction of DNA electrical conductance.
• To reduce computational costs for analyzing DNA charge transport by orders of magnitude.

Main Methods:

• Developed a machine learning model trained on short DNA nanojunctions (3-7 base pairs).
• Used quantum scattering methods to compute electrical conductance for training data, capturing charge-nuclei scattering.
• Input features were designed to represent DNA sequence characteristics.

Main Results:

• The ML model accurately predicts electrical conductance for diverse double-stranded DNA (dsDNA) junctions.
• The model successfully identifies different charge transport mechanisms: quantum tunneling, ballistic transport, and hopping.
• ML predictions align with physical observations of nucleotide clusters influencing DNA transport.

Conclusions:

• Machine learning offers a computationally inexpensive approach to predict DNA electrical conductance.
• The developed ML model can screen millions of dsDNA sequences, accelerating materials discovery.
• The input features are transferable to other ML studies for complex polymer electronics and thermoelectrics.