arxiv: 2605.07175 · v1 · submitted 2026-05-08 · 💻 cs.LG · cs.AI

Recognition: 2 theorem links

· Lean Theorem

Learning Multi-Relational Graph Representations for DNA Methylation-Based Biological Age Estimation

Qing Qing , Xikun Zhang , Zhongyuan Zhang , Jiarui Liu , Xingtong Yu , Xiaotao Shen , Ziqi Xu , Qixin Zhang

show 2 more authors

Zhe Wang Renqiang Luo

Authors on Pith no claims yet

Pith reviewed 2026-05-11 00:50 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords DNA methylationbiological age estimationgraph neural networksmulti-relational graphsCpG sitesaging clocksage accelerationdisease cohorts

0 comments

The pith

RelAge-GNN models three biological graphs among CpG sites to estimate age with higher accuracy and stronger disease sensitivity than independent-feature baselines.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Most DNA methylation aging clocks treat individual CpG sites as unrelated measurements. This paper constructs three separate graphs that link sites through co-methylation patterns, genomic proximity, and gene associations, then runs each graph through its own GNN branch before a gating layer combines the results. The resulting model reports competitive accuracy on large datasets together with tighter correlation to chronological age and greater ability to flag accelerated aging in disease groups. If the relational structure is what drives the gains, then future clocks can embed known biology of site interactions instead of discarding it.

Core claim

By building three complementary graphs that encode co-methylation patterns, genomic co-localization, and gene-level associations among CpG sites, feeding each graph to an independent GNN branch, and fusing the branch outputs with a learnable gating mechanism, RelAge-GNN produces biological age estimates that match or exceed state-of-the-art accuracy while exhibiting stronger correlation with chronological age and improved detection of age acceleration across multiple disease cohorts.

What carries the argument

A multi-relational graph neural network consisting of three independent GNN branches, one per relationship graph, fused by a learnable gating mechanism.

If this is right

The model achieves competitive predictive accuracy and stronger correlation with chronological age than existing methods.
Age acceleration is detected with greater sensitivity across diverse disease cohorts.
Post-hoc analyses quantify the contribution of each relational graph and individual CpG sites to the predictions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same graph-construction strategy could be applied to other epigenetic or multi-omic datasets that contain known relational structure.
Interpretability outputs identifying high-contribution CpG sites may suggest candidate targets for experimental validation of aging interventions.
If the gating weights prove stable across cohorts, they could serve as a compact summary of which biological relationships matter most for aging.

Load-bearing premise

The three constructed graphs sufficiently capture the relevant heterogeneous biological relationships among CpG sites to produce meaningful performance gains over models that treat sites as independent features.

What would settle it

An ablation experiment on the same large-scale datasets in which removing any one of the three graphs leaves accuracy and correlation unchanged or improved would show that the multi-relational construction is not responsible for the reported gains.

Figures

Figures reproduced from arXiv: 2605.07175 by Jiarui Liu, Qing Qing, Qixin Zhang, Renqiang Luo, Xiaotao Shen, Xikun Zhang, Xingtong Yu, Zhe Wang, Zhongyuan Zhang, Ziqi Xu.

**Figure 1.** Figure 1: The framework of RelAge-GNN. 4 Methodology In this section, we present the proposed RelAge-GNN model for age prediction using three-graph methylation data. This model is designed for DNA methylation-based age estimation and extends the traditional single-graph modeling approach to jointly model three types of biological relationships: co-methylation, same-chromosome, and same-gene relationships. Unlike met… view at source ↗

**Figure 2.** Figure 2: Evaluation of AA as a biomarker and distribution across disease cohorts. Top Row: Left: [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Scatter plots illustrate the correlation between predicted and true age for GraphAge, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Node attribute importance and temporal trends. The left panel displays the importance [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Model performance across age and sex groups. Left: Comparison of MAE across three [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗

**Figure 6.** Figure 6: Temporal dynamics of node importance. Left: Age-related trends for the top [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗

**Figure 7.** Figure 7: Average importance scores of component graphs. This figure illustrates the mean importance [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗

read the original abstract

Aging clocks aim to estimate biological age, a measure of physiological state distinct from chronological age, from observable biomarkers, and are widely used for health assessment and disease analysis. DNA methylation is a particularly informative biomarker due to its stability and strong association with aging, and recent learning-based approaches have improved predictive performance. However, most existing methods treat CpG sites as independent features, overlooking the complex and heterogeneous biological relationships among them. We propose RelAge-GNN, a multi-relational graph neural network framework for DNA methylation-based age prediction. Our method constructs three complementary graphs capturing co-methylation patterns, genomic co-localization, and gene-level associations among CpG sites. Each graph is modeled by an independent GNN branch, and a learnable gating mechanism adaptively fuses the resulting representations. Experiments on large-scale datasets show that RelAge-GNN achieves competitive accuracy and stronger correlation with chronological age compared to state-of-the-art methods. Moreover, the model exhibits improved sensitivity in detecting age acceleration across diverse disease cohorts, highlighting its potential utility for disease characterization. Finally, through post hoc interpretability analyses, we quantify the contributions of different relational structures and CpG sites, providing biologically meaningful insights and suggesting potential directions for aging-related research. Our code is available at: https://anonymous.4open.science/r/RelAge-GNN-F1E3/.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

RelAge-GNN applies three biologically motivated graphs and gating to methylation age prediction, but lacks controls to show the relations drive gains over capacity.

read the letter

The main takeaway is that this paper takes the standard methylation age clock task and adds a multi-relational GNN that builds separate graphs for co-methylation patterns, genomic co-localization, and gene associations, then fuses the branch outputs with a learnable gate. That construction is a direct response to the independent-feature limitation noted in prior work, and the code release plus interpretability section are practical additions that let others inspect which relations matter most. The approach is a reasonable extension of GNN ideas into this biomarker setting and gives a concrete way to encode heterogeneous CpG relationships rather than flattening them into a feature vector. The experiments claim competitive accuracy and stronger disease sensitivity, which aligns with the goal of better health monitoring tools. The soft spots sit in the attribution of results. The stress-test note is on point: without ablations against a plain MLP of matched parameter count or against random graphs of similar density, any reported lift could come from extra model capacity rather than the specific biological edges. Graph construction details are also critical—if co-methylation edges are computed on the full dataset before the train-test split, that introduces leakage that would inflate performance. The abstract mentions large-scale datasets and post-hoc analyses, but the lack of error bars, exact baseline descriptions, or split protocols in the summary leaves the robustness open. This work is aimed at researchers building aging clocks or applying graphs to epigenomic data. Someone already working on biomarker prediction or relational models in biology would find the setup and code useful to try or extend. I would send it for peer review. The modeling choice is clear and the domain is applied enough that targeted revisions on controls and leakage checks could make the contribution solid.

Referee Report

3 major / 2 minor

Summary. The paper introduces RelAge-GNN, a multi-relational graph neural network for DNA methylation-based biological age estimation. It constructs three graphs capturing co-methylation patterns, genomic co-localization, and gene-level associations among CpG sites; processes each via an independent GNN branch; and fuses the representations with a learnable gating mechanism. Experiments on large-scale datasets are reported to yield competitive accuracy and stronger correlation with chronological age versus state-of-the-art methods, plus improved sensitivity to age acceleration in diverse disease cohorts. Post-hoc interpretability analyses quantify contributions of relational structures and CpG sites, and code is released.

Significance. If the performance gains are shown to stem specifically from the biologically motivated multi-relational structure rather than model capacity or leakage, the approach could advance aging-clock methodology by incorporating heterogeneous CpG relationships, potentially improving disease characterization and yielding interpretable biological insights. The public code release is a clear strength supporting reproducibility.

major comments (3)

[§3.1] §3.1 (Graph Construction): The co-methylation graph is built from pairwise correlations across samples. It is not stated whether this computation is restricted to the training partition or performed on the full dataset before splitting; the latter would constitute data leakage that invalidates claims of improved generalization and disease sensitivity.
[§4] §4 (Experiments): No ablation is reported that compares RelAge-GNN against (i) a random graph of matched edge density or (ii) a non-relational baseline (e.g., MLP or single GNN) with identical parameter count and gating. Without these controls, observed gains in correlation and disease-cohort sensitivity cannot be attributed to the specific multi-relational inductive bias rather than added capacity from the three GNN branches.
[§4.3] §4.3 (Results, disease cohorts): The claim of “improved sensitivity in detecting age acceleration” lacks reported statistical tests, confidence intervals, or cross-validation details for the disease-cohort comparisons. This weakens the central assertion that the model exhibits stronger biological utility.

minor comments (2)

[Abstract] The abstract states “competitive accuracy” without any numerical values, baseline names, or dataset sizes; these should be added for immediate readability.
[§3.3] Notation for the gating fusion (Eq. (X)) is introduced without an explicit equation reference in the main text; add a numbered equation and clarify the dimension of the gating vector.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback, which has helped clarify key methodological details and strengthen the empirical support for our claims. We address each major comment below and have revised the manuscript accordingly.

read point-by-point responses

Referee: [§3.1] §3.1 (Graph Construction): The co-methylation graph is built from pairwise correlations across samples. It is not stated whether this computation is restricted to the training partition or performed on the full dataset before splitting; the latter would constitute data leakage that invalidates claims of improved generalization and disease sensitivity.

Authors: We appreciate this critical point on data leakage. The co-methylation graph was constructed using pairwise correlations computed exclusively on the training partition within each cross-validation fold. We have now explicitly documented this procedure, including the per-fold computation details, in the revised §3.1 to eliminate any ambiguity. revision: yes
Referee: [§4] §4 (Experiments): No ablation is reported that compares RelAge-GNN against (i) a random graph of matched edge density or (ii) a non-relational baseline (e.g., MLP or single GNN) with identical parameter count and gating. Without these controls, observed gains in correlation and disease-cohort sensitivity cannot be attributed to the specific multi-relational inductive bias rather than added capacity from the three GNN branches.

Authors: We agree that isolating the contribution of the multi-relational structure requires these controls. We have added the requested ablations in the revised §4: (i) RelAge-GNN with random graphs of matched edge density and (ii) a non-relational MLP baseline with identical parameter count and gating mechanism. The new results demonstrate that performance advantages are retained, supporting the role of the biologically motivated relations. revision: yes
Referee: [§4.3] §4.3 (Results, disease cohorts): The claim of “improved sensitivity in detecting age acceleration” lacks reported statistical tests, confidence intervals, or cross-validation details for the disease-cohort comparisons. This weakens the central assertion that the model exhibits stronger biological utility.

Authors: We thank the referee for noting this reporting gap. The revised §4.3 now includes paired statistical tests (with p-values), confidence intervals, and explicit cross-validation details for all disease-cohort comparisons, confirming the significance of the observed improvements in age-acceleration sensitivity. revision: yes

Circularity Check

0 steps flagged

Standard supervised GNN pipeline with no circular derivations

full rationale

The paper describes a multi-relational GNN model that constructs three graphs from biological relationships and methylation data patterns, processes them via independent GNN branches with gating fusion, and trains in a standard supervised manner to predict age. No equations or steps in the abstract or described pipeline reduce predictions to fitted parameters by construction, invoke self-citations as load-bearing uniqueness theorems, or smuggle ansatzes via prior work. The central claims rest on empirical performance comparisons rather than any self-referential derivation chain. This is the expected non-finding for a typical applied ML architecture paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides insufficient detail to enumerate specific free parameters or invented entities; the framework rests on the domain assumption that relational structure among CpG sites is biologically meaningful and capturable by the three graph types.

axioms (1)

domain assumption CpG sites exhibit meaningful co-methylation, co-localization, and gene-level relationships that improve age prediction when modeled jointly
Invoked by the decision to construct and fuse three graphs rather than treat sites independently.

pith-pipeline@v0.9.0 · 5573 in / 1193 out tokens · 41412 ms · 2026-05-11T00:50:45.511348+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

constructs three complementary graphs capturing co-methylation patterns, genomic co-localization, and gene-level associations... independent GNN branch, and a learnable gating mechanism
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Experiments on large-scale datasets show that RelAge-GNN achieves competitive accuracy

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

40 extracted references · 1 canonical work pages

[1]

How to measure biological aging in humans

Luigi Ferrucci, Nir Barzilai, Daniel W Belsky, et al. How to measure biological aging in humans. Nature Medicine, 31(4):1057, 2025

2025
[2]

DNA methylation entropy is a biomarker for aging.Aging (Albany NY), 17(3):685, 2025

Jonathan Chan, Liudmilla Rubbi, and Matteo Pellegrini. DNA methylation entropy is a biomarker for aging.Aging (Albany NY), 17(3):685, 2025

2025
[3]

Causality-enriched epigenetic age uncouples damage and adaptation.Nature Aging, 4(2):231–246, 2024

Kejun Ying, Hanna Liu, Andrei E Tarkhov, et al. Causality-enriched epigenetic age uncouples damage and adaptation.Nature Aging, 4(2):231–246, 2024

2024
[4]

A robust computational framework for methylation age and disease-risk prediction based on pairwise learning.Nature Computational Science, pages 1–16, 2026

Yu Zhang, Yichen Yao, Yuanhao Tang, et al. A robust computational framework for methylation age and disease-risk prediction based on pairwise learning.Nature Computational Science, pages 1–16, 2026

2026
[5]

DNA methylation age of human tissues and cell types.Genome Biology, 14(10):3156, 2013

Steve Horvath. DNA methylation age of human tissues and cell types.Genome Biology, 14(10):3156, 2013

2013
[6]

A pan-tissue DNA- methylation epigenetic clock based on deep learning.npj Aging, 8(1):4, 2022

Lucas Paulo de Lima Camillo, Louis R Lapierre, and Ritambhara Singh. A pan-tissue DNA- methylation epigenetic clock based on deep learning.npj Aging, 8(1):4, 2022

2022
[7]

DeepMAge: A methylation aging clock developed with deep learning.Aging and Disease, 12(5):1252, 2021

Fedor Galkin, Polina Mamoshina, Kirill Kochetov, et al. DeepMAge: A methylation aging clock developed with deep learning.Aging and Disease, 12(5):1252, 2021

2021
[8]

Reinforced neighborhood selection guided multi-relational graph neural networks.ACM Transactions on Information Systems, 40(4):1–46, 2021

Hao Peng, Ruitong Zhang, Yingtong Dou, et al. Reinforced neighborhood selection guided multi-relational graph neural networks.ACM Transactions on Information Systems, 40(4):1–46, 2021

2021
[9]

GraphAge: Unleashing the power of graph neural network to decode epigenetic aging.PNAS Nexus, 4(6):pgaf177, 2025

Saleh Sakib Ahmed, Nahian Shabab, Abul Hassan Samee, and ohters. GraphAge: Unleashing the power of graph neural network to decode epigenetic aging.PNAS Nexus, 4(6):pgaf177, 2025

2025
[10]

Principal neighbourhood aggregation for graph nets

Gabriele Corso, Luca Cavalleri, Dominique Beaini, et al. Principal neighbourhood aggregation for graph nets. InProceedings of the 34th International Conference on Neural Information Processing Systems, pages 13260–13271, 2020

2020
[11]

GNNExplainer: Generating explanations for graph neural networks

Rex Ying, Dylan Bourgeois, Jiaxuan You, et al. GNNExplainer: Generating explanations for graph neural networks. InProceedings of the 33rd International Conference on Neural Information Processing Systems, pages 9244–9255, 2019

2019
[12]

DNA methylation-based biomarkers and the epigenetic clock theory of ageing.Nature Reviews Genetics, 19(6):371–384, 2018

Steve Horvath and Kenneth Raj. DNA methylation-based biomarkers and the epigenetic clock theory of ageing.Nature Reviews Genetics, 19(6):371–384, 2018

2018
[13]

TIME-seq reduces time and cost of DNA methylation measurement for epigenetic clock construction.Nature Aging, 4(2):261–274, 2024

Patrick T Griffin, Alice E Kane, Alexandre Trapp, et al. TIME-seq reduces time and cost of DNA methylation measurement for epigenetic clock construction.Nature Aging, 4(2):261–274, 2024

2024
[14]

Aging by the clock and yet without a program.Nature Aging, 5(10):1946–1956, 2025

David H Meyer, Alexei A Maklakov, and Björn Schumacher. Aging by the clock and yet without a program.Nature Aging, 5(10):1946–1956, 2025

1946
[15]

ComputAgeBench: Epigenetic aging clocks benchmark

Dmitrii Kriukov, Evgeniy Efimov, Ekaterina Kuzmina, et al. ComputAgeBench: Epigenetic aging clocks benchmark. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 5560–5570, 2025

2025
[16]

Graph learning.F oundations and Trends® in Signal Processing, pages 362–519, 2026

Feng Xia, Ciyuan Peng, Jing Ren, et al. Graph learning.F oundations and Trends® in Signal Processing, pages 362–519, 2026

2026
[17]

Train once and explain everywhere: Pre-training interpretable graph neural networks

Jun Yin, Chaozhuo Li, Hao Yan, et al. Train once and explain everywhere: Pre-training interpretable graph neural networks. InProceedings of the 37th International Conference on Neural Information Processing Systems, pages 35277–35299, 2023

2023
[18]

D4Explainer: In-distribution explanations of graph neural network via discrete denoising diffusion

Jialin Chen, Shirley Wu, Abhijit Gupta, et al. D4Explainer: In-distribution explanations of graph neural network via discrete denoising diffusion. InProceedings of the 37th Conference on Neural Information Processing Systems, pages 78964–78986, 2023. 10

2023
[19]

GraphXAI: A survey of graph neural networks (GNNs) for explainable AI (XAI).Neural Computing and Applications, 37(17):10949– 11000, 2025

Mauparna Nandan, Soma Mitra, and Debashis De. GraphXAI: A survey of graph neural networks (GNNs) for explainable AI (XAI).Neural Computing and Applications, 37(17):10949– 11000, 2025

2025
[20]

Gated Multimodal Units for Information Fusion

John Arevalo, Thamar Solorio, Manuel Montes-y Gómez, et al. Gated multimodal units for information fusion.arXiv preprint arXiv:1702.01992, 2017

work page Pith review arXiv 2017
[21]

Epigenetic age acceleration was delayed in schizophrenia.Schizophrenia Bulletin, 47(3):803–811, 2021

Xiaohui Wu, Junping Ye, Zhongju Wang, et al. Epigenetic age acceleration was delayed in schizophrenia.Schizophrenia Bulletin, 47(3):803–811, 2021

2021
[22]

Cell-type-specific aging clocks to quantify aging and rejuvenation in neurogenic regions of the brain.Nature Aging, 3(1):121–137, 2023

Matthew T Buckley, Eric D Sun, Benson M George, et al. Cell-type-specific aging clocks to quantify aging and rejuvenation in neurogenic regions of the brain.Nature Aging, 3(1):121–137, 2023

2023
[23]

Aging and environmental exposures alter tissue-specific DNA methylation dependent upon CpG island context.PLoS Genetics, 5(8):e1000602, 2009

Brock C Christensen, E Andres Houseman, Carmen J Marsit, et al. Aging and environmental exposures alter tissue-specific DNA methylation dependent upon CpG island context.PLoS Genetics, 5(8):e1000602, 2009

2009
[24]

Functional relevance of CpG island length for regulation of gene expression.Genetics, 187(4):1077–1083, 2011

Navin Elango and Soojin V Yi. Functional relevance of CpG island length for regulation of gene expression.Genetics, 187(4):1077–1083, 2011

2011
[25]

Modeling transcriptomic age using knowledge-primed artificial neural networks.npj Aging and Mechanisms of Disease, 7(1):15, 2021

Nicholas Holzscheck, Cassandra Falckenhayn, Jörn Söhle, et al. Modeling transcriptomic age using knowledge-primed artificial neural networks.npj Aging and Mechanisms of Disease, 7(1):15, 2021

2021
[26]

The aging transcriptome: Read between the lines.Current Opinion in Neurobiology, 63:170–175, 2020

Anabel Perez-Gomez, Joel N Buxbaum, and Michael Petrascheck. The aging transcriptome: Read between the lines.Current Opinion in Neurobiology, 63:170–175, 2020

2020
[27]

CpG islands–‘a rough guide’.FEBS Letters, 583(11):1713–1720, 2009

Robert S Illingworth and Adrian P Bird. CpG islands–‘a rough guide’.FEBS Letters, 583(11):1713–1720, 2009

2009
[28]

The impact of flanking sequence features on DNA CpG methylation.Compu- tational Biology and Chemistry, 92:107480, 2021

Daniele Santoni. The impact of flanking sequence features on DNA CpG methylation.Compu- tational Biology and Chemistry, 92:107480, 2021

2021
[29]

Making sense of the ageing methy- lome.Nature Reviews Genetics, 23(10):585–605, 2022

Kirsten Seale, Steve Horvath, Andrew Teschendorff, et al. Making sense of the ageing methy- lome.Nature Reviews Genetics, 23(10):585–605, 2022

2022
[30]

Spatial transcriptomic clocks reveal cell proximity effects in brain ageing.Nature, 638(8049):160–171, 2025

Eric D Sun, Olivia Y Zhou, Max Hauptschein, et al. Spatial transcriptomic clocks reveal cell proximity effects in brain ageing.Nature, 638(8049):160–171, 2025

2025
[31]

Proteomic aging clock predicts mortality and risk of common age-related diseases in diverse populations.Nature Medicine, 30(9):2450–2460, 2024

M Austin Argentieri, Sihao Xiao, Derrick Bennett, et al. Proteomic aging clock predicts mortality and risk of common age-related diseases in diverse populations.Nature Medicine, 30(9):2450–2460, 2024

2024
[32]

Machine learning–based brain aging clocks: Tools for deeper insight into neural aging and rejuvenation.Neural Regeneration Research, pages 10–4103, 2026

Andrew Z Ding and Eric D Sun. Machine learning–based brain aging clocks: Tools for deeper insight into neural aging and rejuvenation.Neural Regeneration Research, pages 10–4103, 2026

2026
[33]

An inflammatory aging clock (iAge) based on deep learning tracks multimorbidity, immunosenescence, frailty and cardiovascular aging.Nature Aging, 1(7):598–615, 2021

Nazish Sayed, Yingxiang Huang, Khiem Nguyen, et al. An inflammatory aging clock (iAge) based on deep learning tracks multimorbidity, immunosenescence, frailty and cardiovascular aging.Nature Aging, 1(7):598–615, 2021

2021
[34]

Age and life expectancy clocks based on machine learning analysis of mouse frailty.Nature Communications, 11(1):4618, 2020

Michael B Schultz, Alice E Kane, Sarah J Mitchell, et al. Age and life expectancy clocks based on machine learning analysis of mouse frailty.Nature Communications, 11(1):4618, 2020

2020
[35]

Novel feature selection methods for construction of accurate epigenetic clocks.PLoS Computational Biology, 18(8):e1009938, 2022

Adam Li, Amber Mueller, Brad English, et al. Novel feature selection methods for construction of accurate epigenetic clocks.PLoS Computational Biology, 18(8):e1009938, 2022

2022
[36]

Utility-preserving federated graph learning with dual-perspective fairness.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026

Renqiang Luo, Huafei Huang, Shuo Yu, et al. Utility-preserving federated graph learning with dual-perspective fairness.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026

2026
[37]

Semi-supervised classification with graph convolutional networks

Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InInternational Conference on Learning Representations, 2017. 11

2017
[38]

Inductive representation learning on large graphs

William L Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. InProceedings of the 31st International Conference on Neural Information Processing Systems, pages 1025–1035, 2017

2017
[39]

Graph attention networks

Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. InInternational Conference on Learning Representations, 2018

2018
[40]

Nucleotide distance influences co-methylation between nearby CpG sites.Genomics, 112(1):144–150, 2020

Ornella Affinito, Domenico Palumbo, Annalisa Fierro, et al. Nucleotide distance influences co-methylation between nearby CpG sites.Genomics, 112(1):144–150, 2020. A Related Work A.1 Aging Clock The core goal of the aging clock is to estimate an individual’s biological age using molecular biomark- ers, thereby reflecting the true aging state of the organis...

2020