Soft windowing application to improve analysis of high-throughput phenotyping data.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• High-throughput phenomic projects generate complex datasets with temporal variations.
• Analyzing data from small treatment groups alongside large control groups presents challenges in managing noise from environmental factors.
• There is a need for methods that leverage temporally local controls to maximize analytical power and minimize noise.

Purpose of the Study:

• To introduce 'soft windowing', a novel methodological approach for analyzing phenomic data.
• To improve the identification of genotype-phenotype associations by optimizing the use of control data.
• To enhance the robustness and power of analyses in large-scale phenomic studies.

Main Methods:

• Developed and applied 'soft windowing', an adaptive method that weights control data based on temporal proximity to mutant data.
• Utilized phenotype data from the International Mouse Phenotyping Consortium (IMPC) for method development and validation.
• Compared the soft windowing approach against a standard non-windowed method using resampling and production analysis pipelines.

Main Results:

• Soft windowing demonstrated a 10% reduction in false positives across 2.5 million analyses via resampling.
• The method increased significant P-values by 30% in genotype-phenotype association studies.
• Soft windowing identified more disease models (106 vs. 99) compared to the non-windowed approach using phenotype overlap.

Conclusions:

• Soft windowing is an effective method for analyzing high-throughput phenomic data, improving analytical power and reducing false positives.
• The method is generalizable and applicable to large-scale human phenomic projects, such as UK Biobank and All of Us.
• The R package 'SmoothWin' is available for implementing this approach in research.