A comparison of feature selection methodologies and learning algorithms in the development of a DNA methylation-based telomere length estimator

Doherty, Trevor; Dempster, Emma; Hannon, Eilís; Mill, Jonathan; Poulton, Richie; Corcoran, David L.; Sugden, Karen; Williams, Benjamin; Caspi, Avshalom; Moffitt, Terrie E.; Delany, Sarah Jane; Murphy, Therese M.

doi:10.1101/2022.04.02.486242

Cited by 8 publications

(13 citation statements)

References 85 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…filtering by magnitude of correlation or strength of association) instead of, or in addition to elastic net [53][54][55][56][57][58][59][60]. Whereas the feature pre-selection here required arbitrary decisions on thresholds, other studies have found that feature reduction via PCA optimises DNAm predictors [36,37].…”

Section: Discussionmentioning

confidence: 96%

See 1 more Smart Citation

Refining epigenetic prediction of chronological and biological age

et al. 2023

View full text Add to dashboard Cite

Background Epigenetic clocks can track both chronological age (cAge) and biological age (bAge). The latter is typically defined by physiological biomarkers and risk of adverse health outcomes, including all-cause mortality. As cohort sample sizes increase, estimates of cAge and bAge become more precise. Here, we aim to develop accurate epigenetic predictors of cAge and bAge, whilst improving our understanding of their epigenomic architecture. Methods First, we perform large-scale (N = 18,413) epigenome-wide association studies (EWAS) of chronological age and all-cause mortality. Next, to create a cAge predictor, we use methylation data from 24,674 participants from the Generation Scotland study, the Lothian Birth Cohorts (LBC) of 1921 and 1936, and 8 other cohorts with publicly available data. In addition, we train a predictor of time to all-cause mortality as a proxy for bAge using the Generation Scotland cohort (1214 observed deaths). For this purpose, we use epigenetic surrogates (EpiScores) for 109 plasma proteins and the 8 component parts of GrimAge, one of the current best epigenetic predictors of survival. We test this bAge predictor in four external cohorts (LBC1921, LBC1936, the Framingham Heart Study and the Women’s Health Initiative study). Results Through the inclusion of linear and non-linear age-CpG associations from the EWAS, feature pre-selection in advance of elastic net regression, and a leave-one-cohort-out (LOCO) cross-validation framework, we obtain cAge prediction with a median absolute error equal to 2.3 years. Our bAge predictor was found to slightly outperform GrimAge in terms of the strength of its association to survival (HRGrimAge = 1.47 [1.40, 1.54] with p = 1.08 × 10−52, and HRbAge = 1.52 [1.44, 1.59] with p = 2.20 × 10−60). Finally, we introduce MethylBrowsR, an online tool to visualise epigenome-wide CpG-age associations. Conclusions The integration of multiple large datasets, EpiScores, non-linear DNAm effects, and new approaches to feature selection has facilitated improvements to the blood-based epigenetic prediction of biological and chronological age.

show abstract

Section: Discussionmentioning

confidence: 96%

“…Several studies have highlighted the benefits of feature pre-selection for elastic net [36,37]. Here, we performed preliminary analyses, including differently sized subsets of CpG sites as features in elastic net.…”

Section: Feature Pre-selectionmentioning

confidence: 99%

Refining epigenetic prediction of chronological and biological age

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Several DNAm studies of age and age-related phenotypes have used pre-selection methods (e.g., filtering by magnitude of correlation or strength of association) instead of, or in addition to elastic net 38–45 . Whereas the feature pre-selection here required arbitrary decisions on thresholds, other studies have found that feature reduction via PCA optimises DNAm predictors 46,47 .…”

Section: Discussionmentioning

confidence: 98%

“…Several studies have highlighted the benefits of feature pre-selection for elastic net 46,47 . Here, we performed preliminary analyses, including differently sized subsets of CpG sites as features in elastic net.…”

Section: Feature Pre-selectionmentioning

confidence: 99%

Refining epigenetic prediction of chronological and biological age

Bernabeu

McCartney

Gadd

et al. 2022

Preprint

View full text Add to dashboard Cite

Epigenetic clocks can track both chronological age (cAge) and biological age (bAge). The latter is typically defined by physiological biomarkers and risk of adverse health outcomes, including all-cause mortality. As cohort sample sizes increase, estimates of cAge and bAge become more precise. Here, we aim to refine predictors and improve understanding of the epigenomic architecture of cAge and bAge. First, we perform large-scale (N = 18,413) epigenome-wide association studies (EWAS) of chronological age and all-cause mortality. Next, to improve cAge prediction, we use methylation data from 24,673 participants from the Generation Scotland (GS) study, the Lothian Birth Cohorts (LBC) of 1921 and 1936 and 8 publicly available datasets. Through the inclusion of linear and non-linear age-CpG associations from the EWAS, feature pre-selection/dimensionality reduction in advance of elastic net regression, and a leave-one-cohort-out (LOCO) cross validation framework, we arrive at an improved cAge predictor (median absolute error = 2.3 years across 10 cohorts). In addition, we train a predictor of bAge on 1,214 all-cause mortality events in GS, based on epigenetic surrogates for 109 plasma proteins and the 8 component parts of GrimAge, the current best epigenetic predictor of all-cause mortality. We test this predictor in four external cohorts (LBC1921, LBC1936, the Framingham Heart Study and the Women's Health Initiative study) where it outperforms GrimAge in its association to survival (HRGrimAge = 1.47 [1.40, 1.54] with p = 1.08 x 10-52, and HRbAge = 1.52 [1.44, 1.59] with p = 2.20 x 10-60). Finally, we introduce MethylBrowsR, an online tool to visualize epigenome-wide CpG-age associations.

show abstract

“…There are many types of feature selection including filter, wrapper and embedded methods. Additionality, feature transformation methods such as PCA can be used to reduce the dimensionality of datahowever these methods do not necessarily yield an explicit set of features and can be less valuable in understanding how features drive a model's decision (Doherty et al, 2022). As such, in this study we have adopted a filter method (LGBM feature importance) to compare the impact of each featurethe importance scores can be used to rank each feature.…”

Section: Analysis Of Features Importancementioning

confidence: 99%

Persuasive communication systems: a machine learning approach to predict the effect of linguistic styles and persuasion techniques

Braca

Dondio

2023

JSIT

View full text Add to dashboard Cite

Purpose Prediction is a critical task in targeted online advertising, where predictions better than random guessing can translate to real economic return. This study aims to use machine learning (ML) methods to identify individuals who respond well to certain linguistic styles/persuasion techniques based on Aristotle’s means of persuasion, rhetorical devices, cognitive theories and Cialdini’s principles, given their psychometric profile. Design/methodology/approach A total of 1,022 individuals took part in the survey; participants were asked to fill out the ten item personality measure questionnaire to capture personality traits and the dysfunctional attitude scale (DAS) to measure dysfunctional beliefs and cognitive vulnerabilities. ML classification models using participant profiling information as input were developed to predict the extent to which an individual was influenced by statements that contained different linguistic styles/persuasion techniques. Several ML algorithms were used including support vector machine, LightGBM and Auto-Sklearn to predict the effect of each technique given each individual’s profile (personality, belief system and demographic data). Findings The findings highlight the importance of incorporating emotion-based variables as model input in predicting the influence of textual statements with embedded persuasion techniques. Across all investigated models, the influence effect could be predicted with an accuracy ranging 53%–70%, indicating the importance of testing multiple ML algorithms in the development of a persuasive communication (PC) system. The classification ability of models was highest when predicting the response to statements using rhetorical devices and flattery persuasion techniques. Contrastingly, techniques such as authority or social proof were less predictable. Adding DAS scale features improved model performance, suggesting they may be important in modelling persuasion. Research limitations/implications In this study, the survey was limited to English-speaking countries and largely Western society values. More work is needed to ascertain the efficacy of models for other populations, cultures and languages. Most PC efforts are targeted at groups such as users, clients, shoppers and voters with this study in the communication context of education – further research is required to explore the capability of predictive ML models in other contexts. Finally, long self-reported psychological questionnaires may not be suitable for real-world deployment and could be subject to bias, thus a simpler method needs to be devised to gather user profile data such as using a subset of the most predictive features. Practical implications The findings of this study indicate that leveraging richer profiling data in conjunction with ML approaches may assist in the development of enhanced persuasive systems. There are many applications such as online apps, digital advertising, recommendation systems, chatbots and e-commerce platforms which can benefit from integrating persuasion communication systems that tailor messaging to the individual – potentially translating into higher economic returns. Originality/value This study integrates sets of features that have heretofore not been used together in developing ML-based predictive models of PC. DAS scale data, which relate to dysfunctional beliefs and cognitive vulnerabilities, were assessed for their importance in identifying effective persuasion techniques. Additionally, the work compares a range of persuasion techniques that thus far have only been studied separately. This study also demonstrates the application of various ML methods in predicting the influence of linguistic styles/persuasion techniques within textual statements and show that a robust methodology comparing a range of ML algorithms is important in the discovery of a performant model.

show abstract

A comparison of feature selection methodologies and learning algorithms in the development of a DNA methylation-based telomere length estimator

Cited by 8 publications

References 85 publications

Refining epigenetic prediction of chronological and biological age

Refining epigenetic prediction of chronological and biological age

Refining epigenetic prediction of chronological and biological age

Persuasive communication systems: a machine learning approach to predict the effect of linguistic styles and persuasion techniques

Contact Info

Product

Resources

About