OASIS: An interpretable, finite-sample valid alternative to Pearson's for scientific discovery

Area of Science:

• Statistics
• Genomics
• Computational Biology

Background:

• Contingency tables are widely used in quantitative research but existing statistical tests lack computational efficiency and statistical validity for finite samples.
• Reference-free genomic inference presents a specific challenge requiring novel statistical approaches for analyzing count data.

Approach:

• Developed OASIS (Optimized Adaptive Statistic for Inferring Structure), a novel family of statistical tests for contingency tables.
• Constructed a test statistic linear in the normalized data matrix, enabling closed-form p-value bounds via concentration inequalities.
• Derived the asymptotic distribution of the OASIS test statistic, validating finite-sample bounds and providing interpretability through table decomposition.

Key Points:

• OASIS demonstrates significant power and interpretability in genomic sequencing data analysis.
• The method successfully detects SARS-CoV-2 and Mycobacterium Tuberculosis strains de novo, surpassing current approaches.
• OASIS exhibits robustness to overdispersion common in genomic data, maintaining false discovery rate control where Pearson's chi-squared test fails.

Conclusions:

• OASIS provides a computationally efficient and statistically valid alternative for contingency table analysis.
• Its applications in genomics highlight its potential for novel pathogen detection and robust data analysis.
• OASIS offers superior power in specific scenarios compared to traditional methods like Pearson's chi-squared test.