Revealing low-temperature plasma efficacy through a dose-rate assessment by DNA damage detection combined with machine learning models

Abstract

Low-temperature plasmas have quickly emerged as alternative and unconventional types of radiation that offer great promise for various clinical modalities. As with other types of radiation, the therapeutic efficacy and safety of low-temperature plasmas are ubiquitous concerns, and assessing their dose rates is crucial in clinical settings. Unfortunately, assessing the dose rates by standard dosimetric techniques has been challenging. To overcome this difficulty, we proposed a dose-rate assessment framework that combined the predictive modeling of plasma-induced damage in DNA by machine learning with existing radiation dose-DNA damage correlations. Our results indicated that low-temperature plasmas have a remarkably high dose rate that can be tuned by various process parameters. This attribute is beneficial for inducing radiobiological effects in a more controllable manner.

DNA Strand Breaks and Denaturation as Probes of Chemical Reactivity versus Thermal Effects of Atmospheric Pressure Plasma Jets

“…Similarly to our previous work, 25 we designed several predictive models ( Note S5 ) to determine the extent of possible types of DNA damage for a given combination of APPJ parameter values. As we discussed later, different types of DNA damage showed different trends in their extent as a function of irradiation time.…”

“…To support our experimental findings, we used a predictive model using specific machine learning (ML) algorithms that we applied to predict plasma-induced changes in DNA accurately for a given combination of APPJ parameter values. We designed and implemented the predictive models based on ML algorithms and artificial neural networks (ANNs) previously to model the type and extent of DNA damage . However, our previous work emphasized primarily the evaluation of the plasma dose rate and its changes depending on plasma parameters.…”

“…We designed and implemented the predictive models based on ML algorithms and artificial neural networks (ANNs) previously to model the type and extent of DNA damage. 25 However, our previous work emphasized primarily the evaluation of the plasma dose rate and its changes depending on plasma parameters. What remains unexplored is the possibility of performing plasma jet diagnostics based on the changes induced in the target.…”

“…To accomplish this, we broaden the exciting possibility of ML-based models to predict DNA denaturation, thus to determine APPJ parameters at which the transition between plasma reactivity and thermal effects can occur. Therefore, in contrast to the previous work, 25 we here investigated the prospect of revealing the correlation between DNA damage and plasma properties that can serve as plasma diagnostics, i.e., plasma gas temperature and reactive species formation, that was also supported by predictive modeling of DNA denaturation.…”

A neural network-based PDE solving algorithm with high precision

“…In recent years, deep neural network (DNN), which reveals superior capability to process and to predict complicated systems, has been widely employed in a variety of fields, e.g., natural language processing 5 , computer vision 6 , etc. 7 , 8 Due to its high computational efficiency and scalability, especially on heterogeneous platforms, DNN has become a promising technique in scientific computing and even provides the possibility for real-time PDE solving 9 . Ray et al 10 proposed a neural network-based indicator to correct the irregular solution in the discontinuous Galerkin scheme.…”