Volume I, Issue 1, 2026
-
Prioritizing human pro-longevity genes is challenging because ageing-related evidence is heterogeneous and distributed across multiple biological resources. This study presents a multimodal machine-learning framework for integrating ageing-related gene information and prioritizing candidate human pro-longevity targets. Data were compiled from GenAge, CellAge, GenDR, and ageing-related gene-expression signatures, producing a harmonised dataset of 4,008 unique genes and 2,327 labelled observations. Genes were classified as pro-longevity or anti-longevity/neutral, and the data were divided into stratified training, validation, and computationally held-out test sets using a 70:15:15 split.
Three complementary models, XGBoost, FT-Transformer, and Graph Attention Network v2 (GATv2), were developed and evaluated, with an equal-weight ensemble assessed on the validation set. The ensemble achieved an area under the receiver operating characteristic curve (AUROC) of 0.9751, an area under the precision–recall curve (AUPRC) of 0.9862, a Matthews correlation coefficient (MCC) of 0.8290, and an accuracy of 0.93. On the computationally held-out test set, the XGBoost and FT-Transformer combination achieved an AUROC of 0.9664, an AUPRC of 0.9767, an MCC of 0.8030, and an accuracy of 0.91. These findings demonstrate the potential of multimodal machine learning to prioritize candidate pro-longevity genes, while independent biological validation remains necessary.
Article info: Received: 2 August, 2026 Revised: 31 August, 2026 Accepted: 25 September, 2026 Published: 30 September, 2026
Copyright: © The Author(s) 2026. Published by RITHA Publishing. This article is distributed under the terms and conditions of the license Creative Commons Attribution (CC BY) license
