Scholarlink

Background <p>X-linked Adrenoleukodystrophy (X-ALD) is a rare neurogenetic disorder with highly variable clinical outcomes, making early prediction difficult. Single-omics analyses capture only limited aspects of its biology, motivating exploratory multiomics approaches.</p> Methods <p>We explored a multiomics framework incorporating radiomics, transcriptomics, epigenomics, proteomics, and lipidomics to examine modality specific signals associated with X-ALD. All datasets were sourced independently and lacked overlapping individuals, so each omics layer was modeled separately. The fusion component is therefore presented as a conceptual architecture rather than an empirically fused predictor. SHAP was used for interpretability, and a Variational GAN was applied to expand the limited radiomics data typical of rare disorders.</p> Results <p>Transcriptomics achieved the highest performance (0.99), reflecting its function as a variant pathogenicity classifier trained on ClinVar rather than a direct diagnostic model for X-ALD. Proteomics (0.96) and lipidomics (0.82) also showed strong results, although these were partly driven by the use of <i>p</i>-value derived targets and features and should be interpreted as proof-of-concept. Radiomics (0.75) captured moderate structural differences, while epigenomics (0.56) showed limited discriminative signal due to small sample size and the complex behavior of methylation in neurodegeneration.</p> Conclusion <p>The framework provides an initial multi-layered view of X-ALD, highlighting transcriptomics and proteomic contributions while clarifying practical constraints imposed by non-overlapping cohorts. The study demonstrates the potential of multiomics exploration and emphasizes the need for harmonized, same-subject datasets to enable true integrated prediction in future work.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploration of multi-omics machine learning framework for predicting X-linked Adrenoleukodystrophy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Chakresh Kumar Jain,
  • Kavya Raj Thekkedathu,
  • Sarita Maurya

摘要

Background

X-linked Adrenoleukodystrophy (X-ALD) is a rare neurogenetic disorder with highly variable clinical outcomes, making early prediction difficult. Single-omics analyses capture only limited aspects of its biology, motivating exploratory multiomics approaches.

Methods

We explored a multiomics framework incorporating radiomics, transcriptomics, epigenomics, proteomics, and lipidomics to examine modality specific signals associated with X-ALD. All datasets were sourced independently and lacked overlapping individuals, so each omics layer was modeled separately. The fusion component is therefore presented as a conceptual architecture rather than an empirically fused predictor. SHAP was used for interpretability, and a Variational GAN was applied to expand the limited radiomics data typical of rare disorders.

Results

Transcriptomics achieved the highest performance (0.99), reflecting its function as a variant pathogenicity classifier trained on ClinVar rather than a direct diagnostic model for X-ALD. Proteomics (0.96) and lipidomics (0.82) also showed strong results, although these were partly driven by the use of p-value derived targets and features and should be interpreted as proof-of-concept. Radiomics (0.75) captured moderate structural differences, while epigenomics (0.56) showed limited discriminative signal due to small sample size and the complex behavior of methylation in neurodegeneration.

Conclusion

The framework provides an initial multi-layered view of X-ALD, highlighting transcriptomics and proteomic contributions while clarifying practical constraints imposed by non-overlapping cohorts. The study demonstrates the potential of multiomics exploration and emphasizes the need for harmonized, same-subject datasets to enable true integrated prediction in future work.