Positive-Unlabelled Learning to Identify New Genes Associated with Dietary Restriction

EasyChair Preprint 15729

Abstract

Dietary Restriction (DR) is a popular anti-ageing approach, and Machine Learning (ML) techniques exist to identify DR-related genes. However, these incorrectly label genes with no known evidence as unrelated to DR (negative examples), which reduces their performance. This study presents a new method that employs two-step Positive-Unlabelled Learning (PU), allowing reliable negative example selection. The method trains a classifier to differentiate DR-related and unrelated genes, achieving superior performance (p<0.05) to the non-PU alternative. Thus, we identified four new genes (PRKAB1, PRKAB2, IRS2, PRKAG1) potentially related to DR, supported by existing literature.

Keyphrases: Aprendizaje Positivo Sin Etiquetas, Aprendizaje automático, Bioinformática, Genética del Envejecimiento, Restricción Dietética

@booklet{EasyChair:15729,
  author    = {Jorge Paz-Ruza and Alex A. Freitas and Amparo Alonso-Betanzos and Bertha Guijarro-Berdiñas},
  title     = {Positive-Unlabelled Learning to Identify New Genes Associated with Dietary Restriction},
  howpublished = {EasyChair Preprint 15729},
  year      = {EasyChair, 2025}}