Bridging Paradigms: Hybrid Mechanistic-Discriminative Predictive Models

Doyle, Orla; Tsaneva-Atansaova, K.; Harte, James M.; Tiffin, Paul A.; Tiňo, Peter; Díaz‐Zuccarini, Vanessa

doi:10.1109/tbme.2013.2244598

Cited by 12 publications

(11 citation statements)

References 73 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…only the inputs and outputs are available, the internal processes are known) and instead move towards models that are interpretable and tailored to the particular question and perhaps incorporate mechanistic aspects of the mode of action of the drug (Doyle et al 2013c). Development of methods that are robust to missing data and data acquired using different protocols or scanners are also an important consideration.…”

Section: Discussionmentioning

confidence: 99%

“…How does a compound differ from and/or resemble existing compounds? To answer these questions, we can build models that range from agnostic (no prior hypothesis) to models that are tailored to the question and incorporate mechanistic information (Doyle et al 2013c). This flexibility enables the exploration of the data to answer a particular question and where appropriate incorporate prior knowledge which may uncover previously unknown associations and thus ultimately contribute to hypothesis generation (Oquendo et al 2012).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

The role of machine learning in neuroimaging for drug discovery and development

2015

View full text Add to dashboard Cite

Neuroimaging has been identified as a potentially powerful probe for the in vivo study of drug effects on the brain with utility across several phases of drug development spanning preclinical and clinical investigations. Specifically, neuroimaging can provide insight into drug penetration and distribution, target engagement, pharmacodynamics, mechanistic action and potential indicators of clinical efficacy. In this review, we focus on machine learning approaches for neuroimaging which enable us to make predictions at the individual level based on the distributed effects across the whole brain. Crucially, these approaches can be trained on data from one study and applied to an independent study and, unlike group-level statistics, can be readily use to assess the generalisability to unseen data. In this review, we present examples and suggestions for how machine learning could help answer fundamental questions spanning the drug discovery pipeline: (1) Who should I recruit for this study? (2) What should I measure and when should I measure it? (3) How does the pharmacological agent behave using an experimental medicine model?, and (4) How does a compound differ from and/or resemble existing compounds? Specifically, we present studies from the literature and we suggest areas for the focus of future development. Further refinement and tailoring of machine learning techniques may help realise their tremendous potential for drug discovery and drug validation.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

The role of machine learning in neuroimaging for drug discovery and development

2015

View full text Add to dashboard Cite

show abstract

“…The advantages of a model-based approach to (un)supervised learning and single-subject predictions have been highlighted by several recent papers (Doyle et al, 2013;Wiecki et al, 2015)). In the following, we summarise a few examples of how it has found application in recent neuroimaging studies of patients.…”

Section: Accepted Manuscriptmentioning

confidence: 99%

Computational neuroimaging strategies for single patient predictions

et al. 2017

View full text Add to dashboard Cite

Neuroimaging increasingly exploits machine learning techniques in an attempt to achieve clinically relevant single-subject predictions. An alternative to machine learning, which tries to establish predictive links between features of the observed data and clinical variables, is the deployment of computational models for inferring on the (patho)physiological and cognitive mechanisms that generate behavioural and neuroimaging responses. This paper discusses the rationale behind a computational approach to neuroimaging-based single-subject inference, focusing on its potential for characterising disease mechanisms in individual subjects and mapping these characterisations to clinical predictions. Following an overview of two main approaches - Bayesian model selection and generative embedding - which can link computational models to individual predictions, we review how these methods accommodate heterogeneity in psychiatric and neurological spectrum disorders, help avoid erroneous interpretations of neuroimaging data, and establish a link between a mechanistic, model-based approach and the statistical perspectives afforded by machine learning.

show abstract

“…Apart from potential secular trends and cohort effects of a purely cross-sectional design (for a more detailed discussion of this point see Ziegler et al, 2012a) we here make the strong assumption that the hidden causes of individual differences in elderly are fully captured by the above considered covariate space

scriptD

. As recently pointed out by Doyle et al (2013b), personalized modeling approaches are required in order to make personalized medicine reality. We might speculate that using a sufficiently high-dimensional multivariate parametrization of individual differences including genes, education, cognitive scores (see e.g.…”

Section: Discussionmentioning

confidence: 99%

Individualized Gaussian process-based prediction and detection of local and global gray matter abnormalities in elderly subjects

et al. 2014

View full text Add to dashboard Cite

Structural imaging based on MRI is an integral component of the clinical assessment of patients with potential dementia. We here propose an individualized Gaussian process-based inference scheme for clinical decision support in healthy and pathological aging elderly subjects using MRI. The approach aims at quantitative and transparent support for clinicians who aim to detect structural abnormalities in patients at risk of Alzheimer's disease or other types of dementia. Firstly, we introduce a generative model incorporating our knowledge about normative decline of local and global gray matter volume across the brain in elderly. By supposing smooth structural trajectories the models account for the general course of age-related structural decline as well as late-life accelerated loss. Considering healthy subjects' demography and global brain parameters as informative about normal brain aging variability affords individualized predictions in single cases. Using Gaussian process models as a normative reference, we predict new subjects' brain scans and quantify the local gray matter abnormalities in terms of Normative Probability Maps (NPM) and global z-scores. By integrating the observed expectation error and the predictive uncertainty, the local maps and global scores exploit the advantages of Bayesian inference for clinical decisions and provide a valuable extension of diagnostic information about pathological aging. We validate the approach in simulated data and real MRI data. We train the GP framework using 1238 healthy subjects with ages 18–94 years, and predict in 415 independent test subjects diagnosed as healthy controls, Mild Cognitive Impairment and Alzheimer's disease.

show abstract

Bridging Paradigms: Hybrid Mechanistic-Discriminative Predictive Models

Cited by 12 publications

References 73 publications

The role of machine learning in neuroimaging for drug discovery and development

The role of machine learning in neuroimaging for drug discovery and development

Computational neuroimaging strategies for single patient predictions

Individualized Gaussian process-based prediction and detection of local and global gray matter abnormalities in elderly subjects

Contact Info

Product

Resources

About