临床试验中心-医疗器械临床研究-临床前实验-苏州普瑞克医疗科技有限公司

SEARCH

与我们合作

普瑞克是美凤力集团下属一家医疗器械合同研究组织(CRO),专注于医疗器械临床研究领域
主要业务为医疗器械全生命周期的临床研究服务、医疗器械合规准入的法规服务。

您也可通过下列途径与我们取得联系:

地 址: 公司总部位于苏州,分布在北京、广州、武汉等地

手 机: 18970915868(胡经理)

邮 箱: chenyajun@meifengli.com

快速提交您的需求 ↓

Perspectives of data science in preclinical safety assessment

更新时间:2024-07-08
查看:33

        The data landscape in preclinical safety assessment is fundamentally changing because of not only emerging new data types, such as human systems biology, or real-world data (RWD) from clinical trials, but also technological advancements in data-processing software and analytical tools based on deep learning approaches. The recent developments of data science are illustrated with use cases for the three factors: predictive safety (new in silico tools), insight generation (new data for outstanding questions); and reverse translation (extrapolating from clinical experience to resolve preclinical questions). Further advances in this field can be expected if companies focus on overcoming identified challenges related to a lack of platforms and data silos and assuring appropriate training of data scientists within the preclinical safety teams.

Introduction

Predictive toxicology aims to use high-quality and reliable scientific data from well-designed in vitro and animal experiments in a structured manner to extrapolate and predict toxicities of untested molecules. Over the past few decades, drug developers and regulatory authorities have increasingly utilized computational (in silico) approaches in combination with data-mining techniques to model toxicological data.1

In particular for simpler in vitro toxicity study types (e.g., in vitro mutagenicity testing or hERG inhibition2, 3), the systematic reuse of data for the purpose of cross-compound comparisons and the generation of predictive tools has led to improved candidate selection and compound optimization.

By contrast, the wealth of high-quality data acquired over years or decades on more complex in vivo study data sets, typically stored for long periods in archives, has been largely undervalued. The economic consequences of high attrition rates as well as increasing ethical considerations regarding animal studies4 have led to a changing view on the value of these legacy data and was one of the triggers for the implementation of the Findability, Accessibility, Interoperability, and Reuse (FAIR) principles.5 In parallel, the development of the Standard for Exchange of Nonclinical Data (SEND) by the Clinical Data Interchange Standards Consortium (CDISC) came into force in December 20166 and resulted in numerous initiatives (e.g., IMI eTRANSAFE and IQ Consortium Non-Clinical to Clinical Translational Database), which try to leverage the new data standard to enhance the reuse of ‘sendified’ legacy data along with new studies.7, 8

However, it also became obvious that the creation of a standard alone does not result in a straightforward, meaningful, and efficient way to scientifically exploit these data for predictive toxicology.9 In addition, tools for storing, querying, reporting, and visualizing the large multidimension data sets are still in their infancy, limiting the versatility of data analysis and identification of insights.

Finally, connecting preclinical safety data with other important data sets from pharmacology (target, off-target data, efficacy results, -omics data, etc.), absorption, distribution, metabolism, and excretion (ADME), or even clinical data poses a major hurdle because of the lack of joint data platforms and missing controlled terminologies (CTs). The limited availability of such cross-sectional data restricts the use cases and, thus, in turn impedes progress in the development of data analysis tools.

Given these shortcomings, here we provide an overview of the current advances in data science covering preclinical development along with a series of use-case examples stratified along three different factors. Furthermore, areas requiring progress in the development of new capabilities and technologies are discussed.

Section snippets

Factors and use cases of data science contributions

We propose three factors to categorize the collected data science use cases based on the questions being answered, the underlying data types, and the analytical challenges to be solved.

The first factor is ‘prediction’, where specific data sets collected in the past (training data) are used to predict effects for new compounds or targets. The selection of the right prediction algorithm suited for the problem, and the training and validation of the corresponding prediction models, as well as

Prediction of mutagenicity and skin sensitization

Given the availability of large amounts of high-quality data2 and a comparably well-characterized mode of action for direct-acting mutagens, tools for predicting the mutagenicity results of Ames test (Salmonella typhimurium reverse mutation assay) were the first to undergo validation.10, 11, 12, 13 This eventually culminated in the first in silico application for regulatory purposes for the assessment and control of DNA reactive impurities.14

Skin sensitization (i.e., an allergic response

Exploring the translation value of preclinical skin findings for the class of kinase inhibitors

The electronic safety platform ToxHub was recently developed under IMI eTRANSAFE7,44 integrating clinical and preclinical data sources and newly developed tools to translate preclinical to clinical terminologies. ToxHub enables sophisticated queries and visualization of complex data, thereby facilitating translational analyses. The preclinical data sources of ToxHub mainly comprise proprietary company study data that were shared among the participants of the project.

To demonstrate the

Polygenic risk for immunogenicity

Treatment with cancer immune checkpoint therapies, specifically PD-1 inhibitors, is associated with severe immune-related adverse events. The underlying risk factors and biological mechanisms are poorly understood and cannot explain the large interpatient variability seen.

Scientists at Genentech53 used a combination of genome-wide association study data and clinical data to identify a polygenic risk score for hypothyroidism in the human population, and applied it to over 2500 patients in the

Challenges and key enablers

The preceding examples have highlighted promising advancements in data science approaches for preclinical development. However, our personal experiences have revealed several key challenges that hinder the widespread adoption of such technologies in the pharmaceutical industry.

Concluding remarks

Further advances in preclinical data science can be expected if companies focus on overcoming the described challenges of the lack of platforms and data silos and assuring appropriate embedding and training of data scientists within the preclinical safety teams, as well as data sharing among industry partners.

Conflict of interest

The authors are employees of the companies listed in the affiliations. They declare that they do not have further conflicts of interest.

Funding

This research received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreements eTRANSAFE (777365). This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA companies in kind contribution.

Acknowledgments

The authors acknowledge the valuable contributions to this review from Eunice Musvasva, Michael Bscheider, Martin Bopst, and Guillemette Duchateau-Nguyen.



QQ客服 电话咨询