Public Abstract

DE-SC0025339: Understanding the mechanism and health consequences of low-dose radiation at a molecular level

Award Status: Active

Institution: Hauptman Woodward Medical Resarch Institute, Inc, Buffalo, NY
UEI: HVQYC8JZB9A6
DUNS:

Most Recent Award Date: 08/16/2024
Number of Support Periods: 1
PM: Kulkarni, Resham

Current Budget Period: 09/01/2024 - 08/31/2025
Current Project Period: 09/01/2024 - 08/31/2027
PI: Snell, Edward

Supplement Budget Period: N/A

Public Abstract

Low-dose and low-dose-rate radiation effects on human health outcomes and the biological mechanisms of these effects are not fully understood but there are concerns that such exposure could affect human health. Structural biology techniques, primarily X-ray crystallography and more recently cryo-electron microscopy, have been overwhelmingly impactful for key discoveries on the molecular basis for health. However, these techniques make use of ionizing radiation or particles and cause high-dose radiation chemistry processes effectively masking any zero-dose state. This study aims to model molecular-level low-dose radiation chemistry, offering insights into its systemic biological impacts. Leveraging a combination of classical machine learning and recent advancements in generative artificial intelligence, we will analyze biological model data across varying dose extremes, integrating additional data and theoretical findings related to low dosage impacts from radiation chemistry literature. Experimental phases will employ engineered proteins to illuminate radiation's structural impacts at low doses, allowing for refinement of machine learning models and validation of computational predictions. Our objective is innovative, using machine learning to develop an artificial intelligence model to understand and predict low-dose radiation health effects at the molecular level and an experimental testing and validation approach to assess confidence in that model. We combine existing but disparate data with both experimental and computational techniques to build the necessary database. Using machine learning to aid the development of an artificial intelligence model, we will obtain an understanding and prediction of the health-related impact of low-dose (10-100 mGy) radiation. We will then extend this understanding of absorbed dose to dose-rate studies. We will incorporate an experimental feedback loop for both dose and dose rate analysis to test, refine, and validate our predictions, providing a measure of confidence in our understanding. Our output will be structural models predicting low-dose damage at a residue and potentially atom-specific level. We will build a training set incorporating experimentally derived structural information from the Protein Data Bank (PDB) and link that to an X-ray dose, or an estimate of dose derived from perturbations seen in that information. Simultaneously, we will select expert literature associated with amino acid and whole protein radiation chemistry across a range of X-ray doses and build this knowledge into our training set. We will then apply machine learning and artificial intelligence to make predictions of low-dose impact. Using the training set, we will develop a computational system that can make predictions of low-dose radiation impact on proteins of interest at a molecular resolution sufficient to provide mechanistic understanding. Experimentally we will probe low dose effects and validate the computational model. We will take two approaches, engineering proteins that undergo large structural changes with radical attack on specific residues, and in parallel, we will produce samples of proteins that are computationally predicted to be easily damaged to provide more experimental knowledge for our computational system. We will use low-dose structural approaches to study these samples to improve and validate the model’s predictive capability and couple this with zero-dose neutron diffraction, and high-dose X-ray diffraction to help understand structural consequences at the residue level. A key finding of The National Academies 2022 consensus study report “Leveraging Advances in Modern Science to Revitalize Low-Dose Radiation Research in the United States” (2022) was that “Radiation biology studies have contributed to the mechanistic understanding of the effects of radiation on molecular pathways and intra- and extracellular processes” and that “The application of novel and developing technologies will enable more precise definition of the cellular and molecular processes that are affected by low-dose and low-dose-rate exposures.” We aim to achieve this definition through the understanding of these processes at the molecular level. The benefits and outcome include a training set allowing machine learning and artificial intelligence to be applied to understanding low-dose health impacts, the development of a tested and validated artificial intelligence model, and the prediction of proteins in the body likely to have health impacts when exposed to low-dose radiation. We expect to be able to predict outcomes at the residue level presenting them in a manner readable by common display software and interpretable by a wide scientific community. The results will create experimentally testable hypotheses and provide a mechanistic understanding of the effects of low-dose radiation on molecular pathways which can then be related to clinical observations. This interdisciplinary approach promises to uncover previously hidden health outcomes of low-dose radiation exposure, guiding future research to enhance health outcomes in this critical area.

Low-dose and low-dose-rate radiation effects on human health outcomes and the biological mechanisms of these effects are not fully understood but there are concerns that such exposure could affect human health. Structural biology techniques, primarily X-ray crystallography and more recently cryo-electron microscopy, have been overwhelmingly impactful for key discoveries on the molecular basis for health. However, these techniques make use of ionizing radiation or particles and cause high-dose radiation chemistry processes effectively masking any zero-dose state. This study aims to model molecular-level low-dose radiation chemistry, offering insights into its systemic biological impacts. Leveraging a combination of classical machine learning and recent advancements in generative artificial intelligence, we will analyze biological model data across varying dose extremes, integrating additional data and theoretical findings related to low dosage impacts from radiation chemistry literature. Experimental phases will employ engineered proteins to illuminate radiation's structural impacts at low doses, allowing for refinement of machine learning models and validation of computational predictions.

Our objective is innovative, using machine learning to develop an artificial intelligence model to understand and predict low-dose radiation health effects at the molecular level and an experimental testing and validation approach to assess confidence in that model. We combine existing but disparate data with both experimental and computational techniques to build the necessary database. Using machine learning to aid the development of an artificial intelligence model, we will obtain an understanding and prediction of the health-related impact of low-dose (10-100 mGy) radiation. We will then extend this understanding of absorbed dose to dose-rate studies. We will incorporate an experimental feedback loop for both dose and dose rate analysis to test, refine, and validate our predictions, providing a measure of confidence in our understanding. Our output will be structural models predicting low-dose damage at a residue and potentially atom-specific level.

We will build a training set incorporating experimentally derived structural information from the Protein Data Bank (PDB) and link that to an X-ray dose, or an estimate of dose derived from perturbations seen in that information. Simultaneously, we will select expert literature associated with amino acid and whole protein radiation chemistry across a range of X-ray doses and build this knowledge into our training set.

We will then apply machine learning and artificial intelligence to make predictions of low-dose impact. Using the training set, we will develop a computational system that can make predictions of low-dose radiation impact on proteins of interest at a molecular resolution sufficient to provide mechanistic understanding.

Experimentally we will probe low dose effects and validate the computational model. We will take two approaches, engineering proteins that undergo large structural changes with radical attack on specific residues, and in parallel, we will produce samples of proteins that are computationally predicted to be easily damaged to provide more experimental knowledge for our computational system. We will use low-dose structural approaches to study these samples to improve and validate the model’s predictive capability and couple this with zero-dose neutron diffraction, and high-dose X-ray diffraction to help understand structural consequences at the residue level.

A key finding of The National Academies 2022 consensus study report “Leveraging Advances in Modern Science to Revitalize Low-Dose Radiation Research in the United States” (2022) was that “Radiation biology studies have contributed to the mechanistic understanding of the effects of radiation on molecular pathways and intra- and extracellular processes” and that “The application of novel and developing technologies will enable more precise definition of the cellular and molecular processes that are affected by low-dose and low-dose-rate exposures.” We aim to achieve this definition through the understanding of these processes at the molecular level.

The benefits and outcome include a training set allowing machine learning and artificial intelligence to be applied to understanding low-dose health impacts, the development of a tested and validated artificial intelligence model, and the prediction of proteins in the body likely to have health impacts when exposed to low-dose radiation. We expect to be able to predict outcomes at the residue level presenting them in a manner readable by common display software and interpretable by a wide scientific community. The results will create experimentally testable hypotheses and provide a mechanistic understanding of the effects of low-dose radiation on molecular pathways which can then be related to clinical observations.

This interdisciplinary approach promises to uncover previously hidden health outcomes of low-dose radiation exposure, guiding future research to enhance health outcomes in this critical area.