Skip to content

poojanV55/Feature-Selection-Using-Bayes-Deep-Learning

Repository files navigation

Feature Selection Using Bayes Deep Learning

This project investigates the efficacy of Bayes Deep Learning (BayesDL) and the TASSEL Mixed Linear Model (MLM) in feature selection for whole-genome SNP data, focusing on their application to two datasets: the TASSEL test data and the Arabidopsis thaliana dataset. The primary objective is to identify significant Single Nucleotide Polymorphisms (SNPs) associated with phenotypic traits, comparing the performance of a deep learning approach against a traditional statistical method.

Overview

BayesDL:

  • Designed for high-dimensional data and complex genetic architectures.
  • Strong capability to detect complex, non-linear associations, particularly in the Arabidopsis dataset.
  • Computationally intensive, requiring substantial resources, making scaling for larger datasets challenging.

TASSEL MLM:

  • Well-suited for handling population structure and genetic relatedness.
  • Provides clear and interpretable results in simpler genetic environments.
  • Effective in initial SNP exploration, especially in datasets with less genetic complexity.

Key Findings

  • BayesDL offers superior performance in complex scenarios but demands more computational resources.
  • TASSEL MLM remains a valuable tool for initial SNP exploration and simpler genetic environments.
  • A hybrid approach, integrating BayesDL’s probabilistic strengths with MLM’s robust population control, could enhance the accuracy and reliability of genotype-phenotype association studies.

Conclusion

This comparative analysis suggests that both methods offer distinct advantages. Integrating BayesDL and TASSEL MLM can significantly advance genomic research and plant breeding strategies, providing comprehensive insights into genetic architecture.

Future Work

Exploring hybrid approaches that combine the non-linear capabilities of BayesDL with the interpretability and population structure control of TASSEL MLM will likely yield even more powerful tools for genomic prediction and association studies.

About

Feature Selection Using Bayes Deep Learning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published