Scholarlink

Nowadays, Deep Learning (DL) has achieved promising results in many areas, including genomic selection (GS). However, DL is a method based on large amounts of data that a single organization studying genomic selection often doesn’t have so much. Federated Learning (FL) is a machine learning setting where multiple entities collaborate in training a shared global model without sharing data. However, FL often performs poorly in non-identically independently distributed (non-IID) settings, so designing an appropriate model is crucial. For FL in pig GS, we propose a method in this study named Federated Genomic Selection (FedGS), which can alleviate the problem of low precision in non-IID settings. We tested the FedGS model on two datasets, including a simulated dataset from three different pig breeding farms and a real dataset from two different pig breeding farms. The experimental results show that joint GS using the FedGS model obtained better prediction accuracy in most cases despite identically independently distributed (IID) or non-IID settings. The source code of FedGS has been packaged into a docker image ( https://hub.docker.com/r/linjie7674/fedgs ) for applications of the GS program.

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

A Federated Learning Approach for Genomic Selection in Pigs

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jie Lin,
  • Wei Zhao,
  • Zhenyang Zhang,
  • Zhe Zhang,
  • Chao Wu

摘要

Nowadays, Deep Learning (DL) has achieved promising results in many areas, including genomic selection (GS). However, DL is a method based on large amounts of data that a single organization studying genomic selection often doesn’t have so much. Federated Learning (FL) is a machine learning setting where multiple entities collaborate in training a shared global model without sharing data. However, FL often performs poorly in non-identically independently distributed (non-IID) settings, so designing an appropriate model is crucial. For FL in pig GS, we propose a method in this study named Federated Genomic Selection (FedGS), which can alleviate the problem of low precision in non-IID settings. We tested the FedGS model on two datasets, including a simulated dataset from three different pig breeding farms and a real dataset from two different pig breeding farms. The experimental results show that joint GS using the FedGS model obtained better prediction accuracy in most cases despite identically independently distributed (IID) or non-IID settings. The source code of FedGS has been packaged into a docker image ( https://hub.docker.com/r/linjie7674/fedgs ) for applications of the GS program.